SDK CV Python bindings API
-
class
metavision_sdk_cv.
ActivityNoiseFilterAlgorithm
(self: metavision_sdk_cv.ActivityNoiseFilterAlgorithm, width: int, height: int, threshold: int) → None Filter that accepts events if a similar event has happened during a certain time window in the past, in the neighborhood of its coordinates.
Builds a new ActivityNoiseFilterAlgorithm object.
- width
Maximum X coordinate of the events in the stream
- height
Maximum Y coordinate of the events in the stream
- threshold
Length of the time window for activity filtering (in us)
-
static
get_empty_output_buffer
() → metavision_sdk_base.EventCDBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.ActivityNoiseFilterAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.ActivityNoiseFilterAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
process_events_
(self: metavision_sdk_cv.ActivityNoiseFilterAlgorithm, events_buf: metavision_sdk_base.EventCDBuffer) → None This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input/output. This should only be used when the number of output events is the same as the number of input events
- events_buf
Buffer of events used as input/output. Its content will be overwritten. It can be converted to a numpy structured array using .numpy()
-
class
metavision_sdk_cv.
AntiFlickerAlgorithm
(self: metavision_sdk_cv.AntiFlickerAlgorithm, width: int, height: int, filter_length: int = 7, min_freq: float = 50.0, max_freq: float = 70.0, diff_thresh_us: int = 1500) → None Algorithm used to remove flickering events given a frequency interval.
- Parameters
width (int) – Sensor’s width
height (int) – Sensor’s height
filter_length (int) – Number of measures of the same period before outputting an event
min_freq (float) – Minimum frequency of the flickering interval
max_freq (float) – Maximum frequency of the flickering interval
diff_thresh_us (unsigned int) – Maximum difference (us) allowed between two consecutive periods to be considered the same.
-
static
get_empty_output_buffer
() → metavision_sdk_base.EventCDBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.AntiFlickerAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.AntiFlickerAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
set_difference_threshold
(self: metavision_sdk_cv.AntiFlickerAlgorithm, diff_thresh: float) → None Sets the difference allowed between two periods to be considered the same.
- diff_thresh
Maximum difference allowed between two successive periods to be considered the same
-
set_filter_length
(self: metavision_sdk_cv.AntiFlickerAlgorithm, filter_length: int) → bool Sets filter’s length.
- filter_length
Number of values in the output median filter
- return
false if value could not be set (invalid value)
-
set_max_freq
(self: metavision_sdk_cv.AntiFlickerAlgorithm, max_freq: float) → bool Sets maximum frequency of the flickering interval.
- note
The value given has to be strictly superior to minimum frequency :max_freq: Maximum frequency of the flickering interval
- return
false if value could not be set (invalid value)
-
set_min_freq
(self: metavision_sdk_cv.AntiFlickerAlgorithm, min_freq: float) → bool Sets minimum frequency of the flickering interval.
- note
The value given has to be strictly inferior to maximum frequency :min_freq: Minimum frequency of the flickering interval
- return
false if value could not be set (invalid value)
-
class
metavision_sdk_cv.
CameraGeometry
(*args, **kwargs) A camera geometry is a mathematical model allowing to map points from world to image plane and vice versa
Overloaded function.
__init__(self: metavision_sdk_cv.CameraGeometry, width: int, height: int, K: numpy.ndarray[numpy.float32], D: numpy.ndarray[numpy.float32]) -> None
__init__(self: metavision_sdk_cv.CameraGeometry, width: int, height: int, P: numpy.ndarray[numpy.float32], IP: numpy.ndarray[numpy.float32], cx: float, cy: float, A: numpy.ndarray[numpy.float32], zoom_factor: float = 1) -> None
-
camera_to_img
(self: metavision_sdk_cv.CameraGeometry, pt_c: buffer, pt_dist_img: buffer) → None Maps a point from the camera’s coordinates system into the distorted image plane.
- pt_c
The 3D point in the camera’s coordinates system
- pt_dist_img
The mapped point in the distorted image plane
-
camera_to_undist_img
(self: metavision_sdk_cv.CameraGeometry, pt_c: buffer, pt_undist_img: buffer) → None Maps a point from the camera’s coordinates system into the undistorted image plane.
- pt_c
The 3D point in the camera’s coordinates system
- pt_undist_img
The mapped point in the undistorted image plane
-
get_distance_to_image_plane
(self: metavision_sdk_cv.CameraGeometry) → float Gets the distance between the camera’s optical center and the undistorted image plane.
-
get_distortion_maps
(self: metavision_sdk_cv.CameraGeometry, mapx: buffer, mapy: buffer) → None
-
get_homography_and_distortion_maps
(self: metavision_sdk_cv.CameraGeometry, H: buffer, mapx: buffer, mapy: buffer) → None
-
get_image_size
(self: metavision_sdk_cv.CameraGeometry) → tuple Gets the sensor’s size, returns a tuple: (width, height)
-
get_img_to_undist_norm_jacobian
(self: metavision_sdk_cv.CameraGeometry, pt_dist_img: buffer, pt_undist_norm: buffer, J: buffer) → None Computes the undistortion function’s jacobian (Row major mode matrix)
- pt_dist_img
The point in the distorted image plane at which the jacobian is computed
- pt_undist_norm
The point in the undistorted normalized image plane
- J
The computed jacobian
-
get_undist_norm_to_img_jacobian
(self: metavision_sdk_cv.CameraGeometry, pt_undist_norm: buffer, pt_dist_img: buffer, J: buffer) → None Computes the distortion function’s jacobian (Row major mode matrix)
- pt_undist_norm
The point in the undistorted normalized image plane at which the jacobian is computed
- pt_dist_img
The point in the distorted image plane
- J
The computed jacobian
-
get_undist_norm_to_undist_img_transform
(self: metavision_sdk_cv.CameraGeometry, m: buffer) → None Gets the transform that maps a point from the undistorted normalized image plane (i.e. Z = 1) into the undistorted image plane (row major mode matrix)
- m
The transform
-
get_undistortion_maps
(self: metavision_sdk_cv.CameraGeometry, mapx: buffer, mapy: buffer) → None
-
img_to_undist_norm
(self: metavision_sdk_cv.CameraGeometry, pt_dist_img: buffer, pt_undist_norm: buffer) → None Maps a point from the distorted image plane into the undistorted normalized image plane.
- pt_dist_img
The point in the distorted image plane
- pt_undist_norm
The mapped point in the undistorted normalized image plane
-
undist_img_to_undist_norm
(self: metavision_sdk_cv.CameraGeometry, pt_undist_img: buffer, pt_undist_norm: buffer) → None Maps a point from the undistorted image plane into the undistorted normalized image plane.
- pt_undist_img
The point in the undistorted image plane
- pt_undist_norm
The mapped point in the undistorted normalized image plane
-
undist_norm_to_dist_norm
(self: metavision_sdk_cv.CameraGeometry, pt_undist_norm: buffer, pt_dist_norm: buffer) → None Maps a point from the undistorted normalized image plane into the distorted normalized image plane.
- pt_undist_norm
The mapped point in the undistorted normalized image plane
- pt_dist_norm
The mapped point in the distorted normalized image plane
-
undist_norm_to_img
(self: metavision_sdk_cv.CameraGeometry, pt_undist_norm: buffer, pt_dist_img: buffer) → None Maps a point from the undistorted normalized image plane into the distorted image plane.
- pt_undist_norm
The point in the undistorted normalized image plane
- pt_dist_img
The mapped point in the distorted image plane
-
undist_norm_to_undist_img
(self: metavision_sdk_cv.CameraGeometry, pt_undist_norm: buffer, pt_undist_img: buffer) → None Maps a point from the undistorted normalized image plane into the normalized image plane.
- pt_undist_norm
The point in the undistorted normalized image plane
- pt_undist_img
The mapped point in the undistorted image plane
-
vector_img_to_undist_norm
(self: metavision_sdk_cv.CameraGeometry, ctr_dist_img: buffer, vec_dist_img: buffer, ctr_undist_norm: buffer, vec_undist_norm: buffer) → None Maps a vector from the distorted image plane into the undistorted normalized image plane.
- ctr_dist_img
The vector’s starting point in the distorted image plane
- vec_dist_img
The vector in the distorted image plane (the vector must be normalized)
- ctr_undist_norm
The vector’s starting point in the undistorted normalized image plane
- vec_undist_norm
The vector in the undistorted normalized image plane
- note
The output vector is normalized
-
vector_undist_norm_to_img
(self: metavision_sdk_cv.CameraGeometry, ctr_undist_norm: buffer, vec_undist_norm: buffer, ctr_dist_img: buffer, vec_dist_img: buffer) → None Maps a vector from the undistorted normalized image plane into the distorted image plane.
- ctr_undist_norm
The vector’s starting point in the undistorted normalized image plane
- vec_undist_norm
The vector in the undistorted normalized image plane (the vector must be normalized)
- ctr_dist_img
The vector’s starting point in the distorted image plane
- vec_dist_img
The mapped vector in the distorted image plane
- note
The output vector is normalized
-
class
metavision_sdk_cv.
DenseFlowFrameGeneratorAlgorithm
(*args, **kwargs) Algorithm used to generate visualization images of dense optical flow streams.
Overloaded function.
__init__(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm, width: int, height: int, maximum_flow_magnitude: float, visualization_method: Metavision::DenseFlowFrameGeneratorAlgorithm::VisualizationMethod, accumulation_policy: Metavision::DenseFlowFrameGeneratorAlgorithm::AccumulationPolicy, resolution_subsampling: int = -1) -> None
__init__(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm, width: int, height: int, maximum_flow_magnitude: float, accumulation_policy: Metavision::DenseFlowFrameGeneratorAlgorithm::AccumulationPolicy) -> None
-
class
AccumulationPolicy
(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm.AccumulationPolicy, value: int) → None Policy for accumulating multiple flow events at a given pixel
Members:
Average
PeakMagnitude
Last
-
property
name
-
property
-
DenseFlowFrameGeneratorAlgorithm
(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm, flow_buf: metavision_sdk_cv.EventOpticalFlowBuffer) → None Processes a buffer of flow events.
- EventIt
Read-Only input event iterator type. Works for iterators over buffers of EventOpticalFlow or equivalent :it_begin: Iterator to the first input event
- it_end
Iterator to the past-the-end event
- note
Successive calls to process_events will accumulate data at each pixel until generate or reset is called.
-
class
VisualizationMethod
(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm.VisualizationMethod, value: int) → None Method to visualize dense flow fields
Members:
DenseColorMap
Arrows
-
property
name
-
property
-
generate
(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm, frame: numpy.ndarray) → None Generates a flow visualization frame.
- frame
Frame that will contain the flow visualization
- allocate
Allocates the frame if true. Otherwise, the user must ensure the validity of the input frame. This is to be used when the data ptr must not change (external allocation, ROI over another cv::Mat, …).
- note
In DenseColorMap mode, the frame will be reset to zero prior to being filled with the flow visualization. In Arrows mode, the flow visualization will be overlaid on top of the input frame. :invalid_argument: if the frame doesn’t have the expected type and geometry
-
generate_legend_image
(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm, legend_frame: numpy.ndarray) → None Generates a legend image for the flow visualization.
- legend_frame
Frame that will contain the flow visualization legend
- square_size
Size of the generated image
- allocate
Allocates the frame if true. Otherwise, the user must ensure the validity of the input frame. This is to be used when the data ptr must not change (external allocation, ROI over another cv::Mat, …) :invalid_argument: if the frame doesn’t have the expected type
-
process_events
(self: metavision_sdk_cv.DenseFlowFrameGeneratorAlgorithm, flow_np: numpy.ndarray[Metavision::EventOpticalFlow]) → None Processes a buffer of flow events.
- EventIt
Read-Only input event iterator type. Works for iterators over buffers of EventOpticalFlow or equivalent :it_begin: Iterator to the first input event
- it_end
Iterator to the past-the-end event
- note
Successive calls to process_events will accumulate data at each pixel until generate or reset is called.
-
class
metavision_sdk_cv.
Event2dFrequencyBuffer
(self: metavision_sdk_cv.Event2dFrequencyBuffer, size: int = 0) → None Constructor
-
numpy
(self: metavision_sdk_cv.Event2dFrequencyBuffer, copy: bool = False) → numpy.ndarray[Metavision::Event2dFrequency<float>] - Copy
if True, allocates new memory and returns a copy of the events. If False, use the same memory
-
resize
(self: metavision_sdk_cv.Event2dFrequencyBuffer, size: int) → None resizes the buffer to the specified size
- size
the new size of the buffer
-
-
class
metavision_sdk_cv.
Event2dFrequencyClusterBuffer
(self: metavision_sdk_cv.Event2dFrequencyClusterBuffer, size: int = 0) → None Constructor
-
numpy
(self: metavision_sdk_cv.Event2dFrequencyClusterBuffer, copy: bool = False) → numpy.ndarray[Metavision::Event2dFrequencyCluster<float>] - Copy
if True, allocates new memory and returns a copy of the events. If False, use the same memory
-
resize
(self: metavision_sdk_cv.Event2dFrequencyClusterBuffer, size: int) → None resizes the buffer to the specified size
- size
the new size of the buffer
-
-
class
metavision_sdk_cv.
Event2dPeriodBuffer
(self: metavision_sdk_cv.Event2dPeriodBuffer, size: int = 0) → None Constructor
-
numpy
(self: metavision_sdk_cv.Event2dPeriodBuffer, copy: bool = False) → numpy.ndarray[Metavision::Event2dPeriod<float>] - Copy
if True, allocates new memory and returns a copy of the events. If False, use the same memory
-
resize
(self: metavision_sdk_cv.Event2dPeriodBuffer, size: int) → None resizes the buffer to the specified size
- size
the new size of the buffer
-
-
class
metavision_sdk_cv.
EventOpticalFlowBuffer
(self: metavision_sdk_cv.EventOpticalFlowBuffer, size: int = 0) → None Constructor
-
numpy
(self: metavision_sdk_cv.EventOpticalFlowBuffer, copy: bool = False) → numpy.ndarray[Metavision::EventOpticalFlow] - Copy
if True, allocates new memory and returns a copy of the events. If False, use the same memory
-
resize
(self: metavision_sdk_cv.EventOpticalFlowBuffer, size: int) → None resizes the buffer to the specified size
- size
the new size of the buffer
-
-
class
metavision_sdk_cv.
FrequencyAlgorithm
(self: metavision_sdk_cv.FrequencyAlgorithm, width: int, height: int, filter_length: int = 7, min_freq: float = 10.0, max_freq: float = 150.0, diff_thresh_us: int = 1500, output_all_burst_events: bool = False) → None Algorithm used to estimate the flickering frequency (Hz) of the pixels of the sensor.
- Parameters
width (int) – Sensor’s width height (int): Sensor’s height filter_length (int): Number of measures of the same period before outputting an event
min_freq (float) – Minimum frequency to output
max_freq (float) – Maximum frequency to output
diff_thresh_us (unsigned int) – Maximum difference (us) allowed between two consecutive periods to be considered the same.
output_all_burst_events (bool) – Whether all the events of a burst must be output or not
-
static
get_empty_output_buffer
() → metavision_sdk_cv.Event2dFrequencyBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.FrequencyAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_cv.Event2dFrequencyBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.FrequencyAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_cv.Event2dFrequencyBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
set_difference_threshold
(self: metavision_sdk_cv.FrequencyAlgorithm, diff_thresh: float) → None Sets the difference allowed between two periods to be considered the same.
- diff_thresh
Maximum difference allowed between two successive periods to be considered the same
-
set_filter_length
(self: metavision_sdk_cv.FrequencyAlgorithm, filter_length: int) → bool Sets filter filter length.
- filter_length
Number of values in the output median filter
- return
false if value could not be set (invalid value)
-
set_max_freq
(self: metavision_sdk_cv.FrequencyAlgorithm, max_freq: float) → bool Sets maximum frequency to output.
- note
The value given has to be > minimum frequency :max_freq: Maximum frequency to output
- return
false if value could not be set (invalid value)
-
set_min_freq
(self: metavision_sdk_cv.FrequencyAlgorithm, min_freq: float) → bool Sets minimum frequency to output.
- note
The value given has to be < maximum frequency :min_freq: Minimum frequency to output
- return
false if value could not be set (invalid value)
-
class
metavision_sdk_cv.
FrequencyClusteringAlgorithm
(self: metavision_sdk_cv.FrequencyClusteringAlgorithm, width: int, height: int, min_cluster_size: int = 1, max_frequency_diff: float = 5.0, max_time_diff: int = 1000, filter_alpha: float = 0.10000000149011612) → None Fequency clustering algorithm. Processes input frequency events and groups them in clusters.
An event belongs to a cluster if it is connected (8-connectivity) to the cluster, its timestamp is within a certain threshold of the last update of the cluster and its frequency is within a certain threshold of the last updated frequency.
The final position of each cluster is a filtered version of the position of the events that get associated to it.
- Parameters
width (int) – Sensor’s width
height (int) – Sensor’s height
min_cluster_size (int) – Minimum size of a cluster to be output (in pixels)
max_frequency_diff (float) – Maximum frequency difference for an input event to be associated to an existing cluster
max_time_diff (int) – Maximum time difference to link an event to an existing cluster
filter_alpha (float) – Filter weight for updating the cluster position with a new event
-
static
get_empty_output_buffer
() → metavision_sdk_cv.Event2dFrequencyClusterBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.FrequencyClusteringAlgorithm, input_np: numpy.ndarray[Metavision::Event2dFrequency<float>], output_buf: metavision_sdk_cv.Event2dFrequencyClusterBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.FrequencyClusteringAlgorithm, input_buf: metavision_sdk_cv.Event2dFrequencyBuffer, output_buf: metavision_sdk_cv.Event2dFrequencyClusterBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
class
metavision_sdk_cv.
PlaneFittingFlowAlgorithm
(self: metavision_sdk_cv.PlaneFittingFlowAlgorithm, width: int, height: int, radius: int = 3, normalized_flow_magnitude: float = 100, min_spatial_consistency_ratio: float = - 1, max_spatial_consistency_ratio: float = - 1, fitting_error_tolerance: int = - 1, neighbor_sample_fitting_fraction: float = 0.30000001192092896) → None This class is an optimized implementation of the dense optical flow approach proposed in Benosman R., Clercq C., Lagorce X., Ieng S. H., & Bartolozzi C. (2013). Event-based visual flow. IEEE transactions on neural networks and learning systems, 25(2), 407-417.
- note
This dense optical flow approach estimates the flow along the edge’s normal, by fitting a plane locally in the time-surface. The plane fitting helps regularize the estimation, but estimated flow results are still relatively sensitive to noise. The algorithm is run for each input event, generating a dense stream of flow events, but making it relatively costly on high event-rate scenes.
- see
TripletMatchingFlowAlgorithm algorithm for a more efficient but more noise sensitive dense optical flow approach.
- see
SparseOpticalFlowAlgorithm algorithm for a flow algorithm based on sparse feature tracking, estimating the full scene motion, staged hence more efficient on high event-rate scenes, but also more complex to tune and dependent on the presence of trackable features in the scene.
-
static
get_empty_output_buffer
() → metavision_sdk_cv.EventOpticalFlowBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.PlaneFittingFlowAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.PlaneFittingFlowAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
class
metavision_sdk_cv.
PlaneFittingFlowEstimator
(self: metavision_sdk_cv.PlaneFittingFlowEstimator, radius: int = 3, enable_flow_normalization: bool = False, min_spatial_consistency_ratio: float = - 1.0, max_spatial_consistency_ratio: float = - 1.0, fitting_error_tolerance: int = - 1, neighbor_sample_fitting_fraction: float = 0.30000001192092896) → None Class computing the flow’s component in the normal direction of an edge moving in a time surface.
The flow is computed by selecting recent timestamp values in a time surface around a given location, fitting a plane to these timestamps using linear least-squares and inferring the flow from the plane’s estimated parameters. This class enables rejecting visual flow estimates based on two quality indicators. The first indicator is the plane fitting error on the timestamps of the timesurface, which is checked to lie within a configured tolerance. The second indicator, denoted spatial consistency, measures the consistency between the radius of the considered neighborhood and the distance covered by the edge during the time period observed in the local timesurface. The visual flow estimates the speed of the local edge and we can calculate the distance covered by the local edge between the timestamp of the oldest event used for plane fitting and the center timestamp. The ratio between this covered distance and the radius of the neighborhood can be seen as a quality indicator for the estimated visual flow, and can be used to reject visual flow estimates when the spatial consistency ratio lies outside a configured range.
Constructor.
- radius
Radius used to select timestamps in a time surface around a given location
- enable_flow_normalization
Flag to indicate if the estimated flow should be normalized
- min_spatial_consistency_ratio
Lower bound of the acceptable range for the spatial consistency ratio quality indicator. Pass a negative value to disable this test.
- max_spatial_consistency_ratio
Upper bound of the acceptable range for the spatial consistency ratio quality indicator. Pass a negative value to disable this test.
- fitting_error_tolerance
Tolerance used to accept visual flow estimates with low enough fitting error. Pass a negative value to disable this test.
- neighbor_sample_fitting_fraction
Fraction used to determine how many timestamps from the timesurface neighborhood are used to fit the plane.
-
get_flow
(self: metavision_sdk_cv.PlaneFittingFlowEstimator, time_surface: metavision_sdk_core.MostRecentTimestampBuffer, x: int, y: int, c: int = 0, time_limit: int = - 1) → tuple Tries to estimate the visual flow at the given location
- time_surface
Input time surface
- x
Abscissa at which the flow is to be estimated
- y
Ordinate at which the flow is to be estimated
- c
Polarity at which timestamps are to be sampled. If the value is -1, the polarity is automatically determined by looking at the most recent timestamp at the given location
- time_limit
Optional parameter that contains the oldest timestamp used during the flow estimation if the estimation has succeeded
- return
tuple (True, vx, vy) if the estimation has succeeded, (False, None, None) otherwise. vx and vy are expressed in pixels/s
-
class
metavision_sdk_cv.
PeriodAlgorithm
(self: metavision_sdk_cv.PeriodAlgorithm, width: int, height: int, filter_length: int = 7, min_period: float = 6500, max_period: float = 100000.0, diff_thresh_us: int = 1500, output_all_burst_events: bool = False) → None Algorithm used to estimate the flickering period of the pixels of the sensor.
output_all_burst_events (bool): Whether all the events of a burst must be output or not
-
static
get_empty_output_buffer
() → metavision_sdk_cv.Event2dPeriodBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.PeriodAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_cv.Event2dPeriodBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.PeriodAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_cv.Event2dPeriodBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
set_difference_threshold
(self: metavision_sdk_cv.PeriodAlgorithm, diff_thresh: float) → None Sets the difference allowed between two periods to be considered the same.
- diff_thresh
Maximum difference allowed between two successive periods to be considered the same
-
set_filter_length
(self: metavision_sdk_cv.PeriodAlgorithm, filter_length: int) → bool Sets filter filter length.
- filter_length
Number of values in the output median filter
- return
false if value could not be set (invalid value)
-
set_max_period
(self: metavision_sdk_cv.PeriodAlgorithm, max_period: float) → bool Sets maximum period to output.
- note
The value max_period has to be larger than the minimum period :max_period: Maximum period to output
- return
false if value could not be set (invalid value)
-
set_min_period
(self: metavision_sdk_cv.PeriodAlgorithm, min_period: float) → bool Sets minimum period to output.
- note
The value min_period has to be smaller than the maximum period :min_period: Minimum period (us) to output
- return
false if value could not be set (invalid value)
-
static
-
class
metavision_sdk_cv.
RoiMaskAlgorithm
(self: metavision_sdk_cv.RoiMaskAlgorithm, pixel_mask: numpy.ndarray[numpy.float64]) → None Class that only propagates events which are contained in a certain region of interest.
The Region Of Interest (ROI) is defined by a mask (cv::Mat). An event is validated if the mask at the event position stores a positive number.
Alternatively, the user can enable different rectangular regions defined by the upper left corner and the bottom right corner that propagates any event inside them.
Builds a new RoiMaskAlgorithm object which propagates events in the given window.
- pixel_mask
Mask of pixels that should be retained (pixel <= 0 is filtered)
-
enable_rectangle
(self: metavision_sdk_cv.RoiMaskAlgorithm, x0: int, y0: int, x1: int, y1: int) → None Enables a rectangular region defined by the upper left corner and the bottom right corner that propagates any event inside them.
- x0
X coordinate of the upper left corner
- y0
Y coordinate of the upper left corner
- x1
X coordinate of the lower right corner
- y1
Y coordinate of the lower right corner
-
static
get_empty_output_buffer
() → metavision_sdk_base.EventCDBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
max_height
(self: metavision_sdk_cv.RoiMaskAlgorithm) → int Returns the maximum number of pixels (height) of the mask.
- return
Maximum height of the mask
-
max_width
(self: metavision_sdk_cv.RoiMaskAlgorithm) → int Returns the maximum number of pixels (width) of the mask.
- return
Maximum width of the mask
-
pixel_mask
(self: metavision_sdk_cv.RoiMaskAlgorithm) → numpy.ndarray[numpy.float64] Returns the pixel mask of the filter.
- return
cv::Mat containing the pixel mask of the filter
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.RoiMaskAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.RoiMaskAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
process_events_
(self: metavision_sdk_cv.RoiMaskAlgorithm, events_buf: metavision_sdk_base.EventCDBuffer) → None This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input/output. This should only be used when the number of output events is the same as the number of input events
- events_buf
Buffer of events used as input/output. Its content will be overwritten. It can be converted to a numpy structured array using .numpy()
-
set_pixel_mask
(self: metavision_sdk_cv.RoiMaskAlgorithm, mask: numpy.ndarray[numpy.float64]) → None Sets the pixel mask of the filter.
- mask
Pixel mask to be used while filtering
-
class
metavision_sdk_cv.
RotateEventsAlgorithm
(self: metavision_sdk_cv.RotateEventsAlgorithm, width_minus_one: int, height_minus_one: int, rotation: float) → None class that allows to rotate an event stream.
- Note
We assume the rotation to happen with respect to the center of the image
Builds a new RotateEventsAlgorithm object with the given width and height.
- width_minus_one
Maximum X coordinate of the events (width-1)
- height_minus_one
Maximum Y coordinate of the events (height-1)
- rotation
Value in radians used for the rotation
-
static
get_empty_output_buffer
() → metavision_sdk_base.EventCDBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.RotateEventsAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.RotateEventsAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
process_events_
(*args, **kwargs) Overloaded function.
process_events_(self: metavision_sdk_cv.RotateEventsAlgorithm, events_np: numpy.ndarray[metavision_sdk_base._EventCD_decode]) -> None
This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input/output. This method should only be used when the number of output events is the same as the number of input events
- events_np
numpy structured array of events whose fields are (‘x’, ‘y’, ‘p’, ‘t’) used as input/output. Its content will be overwritten
process_events_(self: metavision_sdk_cv.RotateEventsAlgorithm, events_buf: metavision_sdk_base.EventCDBuffer) -> None
This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input/output. This should only be used when the number of output events is the same as the number of input events
- events_buf
Buffer of events used as input/output. Its content will be overwritten. It can be converted to a numpy structured array using .numpy()
-
set_rotation
(self: metavision_sdk_cv.RotateEventsAlgorithm, new_angle: float) → None Sets the new rotation angle.
- new_angle
New angle in rad
-
class
metavision_sdk_cv.
SparseFlowFrameGeneratorAlgorithm
(self: metavision_sdk_cv.SparseFlowFrameGeneratorAlgorithm) → None -
add_flow_for_frame_update
(*args, **kwargs) Overloaded function.
add_flow_for_frame_update(self: metavision_sdk_cv.SparseFlowFrameGeneratorAlgorithm, flow_np: numpy.ndarray[Metavision::EventOpticalFlow]) -> None
Stores one motion arrow per centroid (several optical flow events may have the same centroid) in the motion arrow map to be displayed later using the update_frame_with_flow method.
add_flow_for_frame_update(self: metavision_sdk_cv.SparseFlowFrameGeneratorAlgorithm, flow_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
Stores one motion arrow per centroid (several optical flow events may have the same centroid) in the motion arrow map to be displayed later using the update_frame_with_flow method.
-
clear_ids
(self: metavision_sdk_cv.SparseFlowFrameGeneratorAlgorithm) → None
-
update_frame_with_flow
(self: metavision_sdk_cv.SparseFlowFrameGeneratorAlgorithm, display_mat: numpy.ndarray) → None Updates the input frame with the centroids’ motion stored in the history.
Clears the history afterwards
-
-
class
metavision_sdk_cv.
SparseOpticalFlowAlgorithm
(*args, **kwargs) Overloaded function.
__init__(self: metavision_sdk_cv.SparseOpticalFlowAlgorithm, width: int, height: int, config: metavision_sdk_cv.SparseOpticalFlowConfigPreset = <SparseOpticalFlowConfigPreset.FastObjects: 1>) -> None
__init__(self: metavision_sdk_cv.SparseOpticalFlowAlgorithm, width: int, height: int, distance_gain: float = 0.05000000074505806, damping: float = 0.7070000171661377, omega_cutoff: float = 7.0, min_cluster_size: int = 7, max_link_time: int = 30000, match_polarity: bool = True, use_simple_match: bool = True, full_square: bool = True, last_event_only: bool = False, size_threshold: int = 100000000) -> None
-
static
get_empty_output_buffer
() → metavision_sdk_cv.EventOpticalFlowBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.SparseOpticalFlowAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.SparseOpticalFlowAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
class
metavision_sdk_cv.
SpatioTemporalContrastAlgorithm
(self: metavision_sdk_cv.SpatioTemporalContrastAlgorithm, width: int, height: int, threshold: int, cut_trail: bool = True) → None The SpatioTemporalContrast Filter is a noise filter using the exponential response of a pixel to a change of light to filter out wrong detections and trails.
For an event to be forwarded, it needs to be preceded by another one in a given time window, this ensures that the spatio temporal contrast detection is strong enough. It is also possible to then cut all the following events up to a change of polarity in the stream for that particular pixel (strong trail removal). Note that this will remove signal if 2 following edges of the same polarity are detected (which should not happen that frequently).
- note
The timestamp may be stored in different types 64 bits, 32 bits or 16 bits. The behavior may vary from one size to the other since the number of significant bits may change. Before using the version with less than 32 bits check that the behavior is still valid for the usage.
Builds a new SpatioTemporalContrast object.
- width
Maximum X coordinate of the events in the stream
- height
Maximum Y coordinate of the events in the stream
- threshold
Length of the time window for filtering (in us)
- cut_trail
If true, after an event goes through, it removes all events until change of polarity
-
static
get_empty_output_buffer
() → metavision_sdk_base.EventCDBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.SpatioTemporalContrastAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.SpatioTemporalContrastAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
process_events_
(self: metavision_sdk_cv.SpatioTemporalContrastAlgorithm, events_buf: metavision_sdk_base.EventCDBuffer) → None This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input/output. This should only be used when the number of output events is the same as the number of input events
- events_buf
Buffer of events used as input/output. Its content will be overwritten. It can be converted to a numpy structured array using .numpy()
-
class
metavision_sdk_cv.
TimeGradientFlowAlgorithm
(self: metavision_sdk_cv.TimeGradientFlowAlgorithm, width: int, height: int, radius: int, min_flow_mag: float, bit_cut: int) → None This class is a local and dense implementation of Optical Flow from events.
- note
This approach is dense in the sense that it processes events at the sensor resolution and produces OpticalFlowEvents potentially on the whole sensor matrix. It computes the optical flow along the edge’s normal by analyzing the recent timestamps at only the left, right, top and down K-pixel far neighbors (i.e. not the whole neighborhood). Thus, the estimated flow results are still quite sensitive to noise. The algorithm is run for each input event, generating a dense stream of flow events, but making it relatively costly on high event-rate scenes. The bit size of the timestamp representation can be reduced to accelerate the processing. :timestamp_type: Type of the timestamp used in to compute the optical flow. Typically Metavision::timestamp. Can be used with lighter type (like std::uint32_t) to lower processing time when critical.
-
static
get_empty_output_buffer
() → metavision_sdk_cv.EventOpticalFlowBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.TimeGradientFlowAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.TimeGradientFlowAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
class
metavision_sdk_cv.
TrailFilterAlgorithm
(self: metavision_sdk_cv.TrailFilterAlgorithm, width: int, height: int, threshold: int) → None Filter that accepts an event either if the last event at the same coordinates was of different polarity, or if it happened at least a given amount of time after the last event.
Builds a new TrailFilterAlgorithmT object.
- width
Maximum X coordinate of the events in the stream
- height
Maximum Y coordinate of the events in the stream
- threshold
Length of the time window for activity filtering (in us)
-
static
get_empty_output_buffer
() → metavision_sdk_base.EventCDBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.TrailFilterAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.TrailFilterAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
process_events_
(self: metavision_sdk_cv.TrailFilterAlgorithm, events_buf: metavision_sdk_base.EventCDBuffer) → None This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input/output. This should only be used when the number of output events is the same as the number of input events
- events_buf
Buffer of events used as input/output. Its content will be overwritten. It can be converted to a numpy structured array using .numpy()
-
class
metavision_sdk_cv.
TransposeEventsAlgorithm
(self: metavision_sdk_cv.TransposeEventsAlgorithm) → None Class that switches X and Y coordinates of an event stream. This filter changes the dimensions of the corresponding frame (width and height are switched)
-
static
get_empty_output_buffer
() → metavision_sdk_base.EventCDBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.TransposeEventsAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.TransposeEventsAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_base.EventCDBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
-
process_events_
(*args, **kwargs) Overloaded function.
process_events_(self: metavision_sdk_cv.TransposeEventsAlgorithm, events_np: numpy.ndarray[metavision_sdk_base._EventCD_decode]) -> None
This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input/output. This method should only be used when the number of output events is the same as the number of input events
- events_np
numpy structured array of events whose fields are (‘x’, ‘y’, ‘p’, ‘t’) used as input/output. Its content will be overwritten
process_events_(self: metavision_sdk_cv.TransposeEventsAlgorithm, events_buf: metavision_sdk_base.EventCDBuffer) -> None
This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input/output. This should only be used when the number of output events is the same as the number of input events
- events_buf
Buffer of events used as input/output. Its content will be overwritten. It can be converted to a numpy structured array using .numpy()
-
static
-
class
metavision_sdk_cv.
TripletMatchingFlowAlgorithm
(*args, **kwargs) This class implements the dense optical flow approach proposed in Shiba S., Aoki Y., & Gallego G. (2022). "Fast Event-Based Optical Flow Estimation by Triplet Matching". IEEE Signal Processing Letters, 29, 2712-2716.
- note
This dense optical flow approach estimates the flow along the edge’s normal, by locally searching for aligned events triplets. The flow is estimated by averaging all aligned event triplets found, which helps regularize the estimates, but results are still relatively sensitive to noise. The algorithm is run for each input event, generating a dense stream of flow events, but making it relatively costly on high event-rate scenes.
- see
PlaneFittingFlowAlgorithm algorithm for slightly more accurate but more expensive dense optical flow approach.
- see
SparseOpticalFlowAlgorithm algorithm for a flow algorithm based on sparse feature tracking, estimating the full scene motion, staged hence more efficient on high event-rate scenes, but also more complex to tune and dependent on the presence of trackable features in the scene.
Overloaded function.
__init__(self: metavision_sdk_cv.TripletMatchingFlowAlgorithm, width: int, height: int, radius: float, dt_min: int, dt_max: int) -> None
__init__(self: metavision_sdk_cv.TripletMatchingFlowAlgorithm, width: int, height: int, radius: float, min_flow_mag: float, max_flow_mag: float) -> None
-
static
get_empty_output_buffer
() → metavision_sdk_cv.EventOpticalFlowBuffer This function returns an empty buffer of events of the correct type, which can later on be used as output_buf when calling process_events()
-
process_events
(*args, **kwargs) Overloaded function.
process_events(self: metavision_sdk_cv.TripletMatchingFlowAlgorithm, input_np: numpy.ndarray[metavision_sdk_base._EventCD_decode], output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes a numpy array as input and writes the results into the specified output event buffer
- input_np
input chunk of events (numpy structured array whose fields are (‘x’, ‘y’, ‘p’, ‘t’). Note that this order is mandatory)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()
process_events(self: metavision_sdk_cv.TripletMatchingFlowAlgorithm, input_buf: metavision_sdk_base.EventCDBuffer, output_buf: metavision_sdk_cv.EventOpticalFlowBuffer) -> None
- This method is used to apply the current algorithm on a chunk of events. It takes an event buffer as input and writes the results into a distinct output event buffer
- input_buf
input chunk of events (event buffer)
- output_buf
output buffer of events. It can be converted to a numpy structured array using .numpy()