SDK CV Configs

struct Metavision::FrequencyClusteringAlgorithmConfig

Data needed to configure a FrequencyClusteringAlgorithm.

Public Functions

inline FrequencyClusteringAlgorithmConfig(int min_cluster_size, float max_frequency_diff, timestamp max_time_diff, float filter_alpha)

Constructor.

Parameters

min_cluster_size – Minimum size of a cluster to be output (in pixels)
max_frequency_diff – Maximum frequency difference for an input event to be associated to an existing cluster
max_time_diff – Maximum time difference to link an event to an existing cluster
filter_alpha – Filter weight for updating the cluster position with a new event

FrequencyClusteringAlgorithmConfig() = default: Default constructor.

struct Metavision::FrequencyEstimationConfig

Configuration used when estimating the frequency of periodic signals.

Public Functions

inline FrequencyEstimationConfig(unsigned int filter_length = 7, float min_freq = 10.f, float max_freq = 150.f, unsigned int diff_thresh_us = 1500, bool output_all_burst_events = false)

Constructor.

Note

Internally, the algorithm first estimates period using the events’ timestamps and then converts it to frequency. The decision-making criterion diff_thresh_us is then absolute in the period domain, instead of the frequency domain.

Parameters

filter_length – Number of measures of the same period before outputting an event
min_freq – Minimum frequency to output.
max_freq – Maximum frequency to output.
diff_thresh_us – Maximum difference (us) allowed between two consecutive periods to be considered the same.
output_all_burst_events – Whether all the events of a burst must be output or not

struct Metavision::PeriodEstimationConfig

Configuration used when estimating the period of periodic signals.

Public Functions

inline PeriodEstimationConfig(unsigned int filter_length = 7, float min_period = 6500, unsigned int max_period = 1e5, float diff_thresh_us = 1500, bool output_all_burst_events = false)

Constructor.

Parameters

filter_length – Number of measures of the same period before outputting an event
min_period – Minimum period (us) to output. Default value is 6500us, which corresponds to 153Hz
max_period – Maximum period (us) to output. Default value is 1e5us, which corresponds to 10Hz
diff_thresh_us – Maximum difference (us) allowed between two consecutive periods to be considered the same
output_all_burst_events – Whether all the events of a burst must be output or not

struct Metavision::SparseOpticalFlowConfig

Configuration to use with the sparse optical flow algorithm.

Public Functions

inline SparseOpticalFlowConfig(float distance_gain = 0.05f, float damping = 0.707f, float omega_cutoff = 5.f, unsigned int min_cluster_size = 5, timestamp max_link_time = 50000, bool match_polarity = true, bool use_simple_match = true, bool full_square = true, bool last_event_only = false, unsigned int size_threshold = 100000000)

Constructor.

Parameters

distance_gain – Distance gain of the low pass filter to compute the size of a cluster
damping – Damping parameter of the Luenberger estimator, will determine how fast the speed computation of a cluster will converge. It is related to the eigenvalues of the observer. The damping ratio influences the decay of the oscillations in the observer response. A value of 1 is for critical damping, while a value of 0 produces perfect oscillations. The system is underdamped for values lesser than 1 and overdamped for values greater than 1.
omega_cutoff – Parameter of the Luenberger estimator, will determine how fast the speed computation of a cluster will respond to changes in the system. It is related to the eigenvalues of the observer. A higher frequency results in faster convergence but may introduce more oscillations.
min_cluster_size – Minimal number of events hitting a cluster before it gets outputted
max_link_time – Maximum time in us for two events to be linked in time
match_polarity – Set to true to create mono-polarity clusters, otherwise it will use multi-polarity ones
use_simple_match – Set to false to use a costlier constant velocity match strategy
full_square – Set to true to check connectivity on the full 3x3 square around the events, otherwise it will use the 3x3 cross around it
last_event_only – Set to true to only check the connectivity with the last event added to the cluster referenced at every pixel positions
size_threshold – Threshold on the spatial size of a cluster before being outputted

Public Members

float distance_gain: Gain of the low pass filter to compute the size of a CCL.

float damping: Parameter of the Luenberger estimator, will determine how fast the speed computation of a CCL will converge. The damping ratio influences the decay of the oscillations in the observer response.

float omega_cutoff: Parameter of the Luenberger estimator, will determine how fast the speed computation of a CCL will respond to changes in the system. A higher frequency results in faster convergence but may introduce more oscillations.

unsigned int min_cluster_size: Minimal number of events hitting a cluster before it gets outputted.

timestamp max_link_time: Maximum time difference in us for two events to be linked in time.

bool match_polarity: Decide if we have multi-polarity cluster or mono polarity ones.

bool use_simple_match: If false, will use a (costlier) constant velocity match strategy.

bool full_square: If true, check connectivity on the full 3x3 square around the events, otherwise just the 3x3 cross around it

bool last_event_only: If true will only check the connectivity with the last event added to the cluster referenced at every pixel positions

unsigned int size_threshold: Threshold on the spatial size of a cluster (in pixels). Maximum distance allowed from the estimated center of the cluster to the last event added to it

struct Metavision::TimeGradientFlowAlgorithmConfig

Structure representing the configuration of the time gradient flow algorithm.

Public Functions

inline TimeGradientFlowAlgorithmConfig(uint8_t radius = 3, float min_flow_mag = 0.f, uint8_t bit_cut = 0)

Initializing constructor.

Parameters

radius – Spatial radius for flow analysis
min_flow_mag – Minimum flow magnitude to be observed
bit_cut – Number of bits to remove to compare timestamps (used to accelerate processing)

Public Members

uint8_t local_radius: Matching spatial search radius. The event received will provide the center of the analyzed neighborhood. The pixels at the local_radius distance on the left, right, top and bottom will be used to compute optical flow, not the pixels in-between. The further the distance, the more the flow estimate is regularized but also the more subject to association error.

float min_norm: Minimal flow norm estimated to produce an opticalFlowEvent (the actual value used is 1.f/(min_norm*min_norm) to get a value in pix/s)

uint8_t bit_cut: Number of bits to remove to compare timestamps (used to accelerate processing)

struct Metavision::TripletMatchingFlowAlgorithmConfig

Structure representing the configuration of the triplet matching algorithm.

Public Functions

TripletMatchingFlowAlgorithmConfig() = default: Default constructor.

inline TripletMatchingFlowAlgorithmConfig(float radius, timestamp dt_min, timestamp dt_max)

Initializing constructor.

Parameters

radius – Spatial radius to search for event matches
dt_min – Minimum time difference for event matches
dt_max – Maximum time difference for event matches

inline TripletMatchingFlowAlgorithmConfig(float radius, float min_flow_mag, float max_flow_mag)

Initializing constructor.

Parameters

radius – Spatial radius to search for event matches
min_flow_mag – Minimum flow magnitude to be observed, will be converted to a constraint on matches time difference
max_flow_mag – Maximum flow magnitude to be observed, will be converted to a constraint on matches time difference

Public Members

float radius: Matching spatial search radius. Higher values robustify and regularize estimated visual flow, but increase search area and number of potential matches, hence reduce efficiency. Note that this radius relates to the spatial search for event matching, but that the flow is actually estimated from 2 matches, so potentially twice this area.

timestamp dt_max: Matching temporal upper bound. Higher values lower minimum observable visual flow, but increase search area and number of potential matches, hence reduce efficiency.

timestamp dt_min: Matching temporal lower bound. Higher values reduce influence of noise when searching for matches in the past, by better filtering lateral observations of the same edge, but reduce the maximum observable visual flow.