SDK Analytics Configs

struct JetMonitoringAlarmConfig

Jet monitoring alarm parameters.

Public Functions

inline bool is_valid() const

Basic validation.

Public Members

bool alarm_on_count = false

If true, an alarm will be raised if the jet count exceeds the expected_count value.

int max_expected_count = 0

Maximum expected number of jets.

bool alarm_on_cycle = false

Activate/deactivate alarm on cycle time.

float expected_cycle_ms = 0

Expected cycle time.

float cycle_tol_percentage = 0

Tolerance for estimated cycle time, in percentage of expected_cycle_ms.

struct JetMonitoringAlgorithmConfig

Jet monitoring algorithm parameters.

Public Types

enum Orientation

Enum listing 4 possible nozzle orientations in the image reference frame.

The angle between the nozzle reference frame and the image reference frame must be inside {0; +-90; 180}

Values:

enumerator Down

enumerator Up

enumerator Left

enumerator Right

Public Members

Orientation nozzle_orientation = Down

Nozzle orientation.

cv::Rect detection_roi

Central ROI used to detect jets by identifying peaks in the event-rate.

struct JetMonitoringLoggerConfig

Jet monitoring logger parameters.

Public Members

bool enable_logging = false

If true, the logging is enabled, otherwise no logging.

int log_history_length_ms = 0

Log buffer length in ms. Each log dump will have this duration.

int log_dump_delay_ms = 0

Log delay after the trigger in ms. How much we see after the trigger.

std::string log_out_dir = ""

Base output directory. Each log dump will be a subdirectory in it.

bool dump_at_exit = false

Automatically dump when exiting the application.

bool log_jet_video = false

Combine the images of each jet to create a video of an average jet.

bool log_jets_evt_rate = false

Dump a file containing the event rate for each jet.

struct LineClusterTrackingConfig

Struct representing the parameters used to instantiate a LineClusterTracker inside PsmAlgorithm.

Public Functions

LineClusterTrackingConfig() = default

Default constructor.

inline LineClusterTrackingConfig(unsigned int bitsets_buffer_size, unsigned int cluster_ths = 3, unsigned int num_clusters_ths = 4, unsigned int min_inter_clusters_distance = 1, float learning_rate = 1.f, float max_dx_allowed = 5.f, unsigned int max_nbr_empty_rows = 0)

Constructor.

Parameters

• bitsets_buffer_size – Size of the bitset circular buffer (accumulation_time = bitsets_buffer_size *precision_time_us )

• cluster_ths – Minimum width (in pixels) below which clusters of events are considered as noise

• num_clusters_ths – Minimum number of cluster measurements below which a particle is considered as noise

• min_inter_clusters_distance – Once small clusters have been removed, merge clusters that are closer than this distance. This helps dealing with dead pixels that could cut particles in half. If set to 0, do nothing

• learning_rate – Ratio in the weighted mean between the current x position and the observation. This is used only when the particle is shrinking, because the front of the particle is always sharp while the trail might be noisy. 0.0 is conservative and does not take the observation into account, whereas 1.0 has no memory and overwrites the cluster estimate with the new observation. A value outside ]0,1] disables the weighted mean, and 1.0 is used instead.

• max_dx_allowed – Caps x variation at this value. A negative value disables the clamping. This is used only when the particle is shrinking, because the front of the particle is always sharp while the trail might be noisy.

• max_nbr_empty_rows – Number of consecutive empty measurements that is tolerated

struct LineParticleTrackingConfig

Struct representing the parameters used to instantiate a LineParticleTracker inside PsmAlgorithm.

Public Functions

inline LineParticleTrackingConfig(bool is_going_down, unsigned int dt_first_match_ths, double tan_angle_ths = 1., double matching_ths = 0.5)

Constructor.

Parameters

• is_going_down – True if the particle is falling, false if it’s going upwards

• dt_first_match_ths – Maximum allowed duration to match the 2nd particle of a track

• tan_angle_ths – Tangent of the angle with the vertical beyond which two particles on consecutive lines can’t be matched

• matching_ths – Minimum similarity score in [0,1] needed to match two particles

struct SpatterTrackerAlgorithmConfig

Configuration to instantiate a SpatterTrackerAlgorithmConfig.

Public Functions

inline SpatterTrackerAlgorithmConfig(int cell_width, int cell_height, int untracked_threshold = 5, int activation_threshold = 10, bool apply_filter = true, int max_distance = 50, const int min_size = 1, const int max_size = std::numeric_limits<int>::max(), timestamp min_time_tracked = 0, timestamp static_objects_memory = 0, int max_size_variation = 100, int max_dist_static_obj = 0, EvtFilterType filtered_polarity = EvtFilterType::FILTER_NEG)

Constructor.

Parameters

• cell_width – Width of cells used for clustering

• cell_height – Height of cells used for clustering

• untracked_threshold – Maximum number of times a spatter_cluster can stay untracked before being removed

• activation_threshold – Threshold distinguishing an active cell from inactive cells (i.e. minimum number of events in a cell to consider it as active)

• apply_filter – If true, then a simple filter to remove crazy pixels will be applied

• max_distance – Max distance for clusters association (in pixels)

• min_size – Minimum object size (in pixels) - minimum of the object’s width and height should be larger than this value

• max_size – Maximum object size (in pixels) - maximum of the object’s width and height should be smaller than this value

• min_time_tracked – Minimum delay to track a custer. Before this delay, it is tracked internally only.

• static_objects_memory – Time after which the static constraint of detected clusters is relaxed

• max_size_variation – Maximum size variation allowed for two cluster to be associated

• max_dist_static_obj – Maximum displacement allowed for a static cluster (else, it is considered a moving cluster)

• filtered_polarity – Polarity of events to filter out

Public Members

int cell_width_

Width of the cells used for clustering.

int cell_height_

Height of the cells used for clustering.

int untracked_threshold_

Maximum number of times a spatter cluster can stay untracked before being forgotten.

int activation_threshold_

Minimum number of events in a cell to consider it active during clustering.

bool apply_filter_

If true then will consider only one event per pixel.

int max_distance_

Maximum distance (in pixel) allowed for clusters association.

int min_size_

Minimum object size (in pixel)

int max_size_

Maximum object size (in pixel)

timestamp min_time_tracked_us_

Minimum duration a new cluster should be tracked before being output (in us)

int max_dist_static_obj_

Maximum displacement allowed to consider a cluster static (in pixel)

int max_size_variation_

Maximum size variation allowed for two clusters to be matched (in pixel)

timestamp static_memory_

Time after which the static constraint of the detected clusters is relaxed (in us)

EvtFilterType filter_type_

Polarity of events to filter out.

struct TrackingConfig

Structure used to configure the TrackingAlgorithm.

There are 4 main parts:

1. Clustering using one of the methods indexed by TrackingConfig::ClusterMaker

2. Data association using one of the methods indexed by TrackingConfig::DataAssociation

3. Tracker initialization using the method indexed by TrackingConfig::Initializer

4. Tracking using one of the methods indexed by TrackingConfig::Tracker

Public Types

enum class ClusterMaker

Defines the type of cluster maker used by the TrackingAlgorithm to build clusters from input events.

Values:

enumerator SIMPLE_GRID

Cluster making is based on a regular grid:

• The camera FOV is divided into elementary cells using a regular grid; the size of the cells is cell_width_ x cell_height_.

• For each cell, if the number of events in the cell for the given time-slice cell_deltat_ is larger than the activation_threshold_, then the cell is considered as active.

• Active cells of a given time-slice are connected into clusters.

enumerator MEDOID_SHIFT

Cluster making is based on spatial and temporal distances between neighboring events. If the spatial and temporal distances between the event and its neighboring event are smaller than the thresholds medoid_shift_spatial_dist_ and medoid_shift_temporal_dist_, then the event goes to the same cluster as its neighbour, otherwise, it creates a new cluster.

enum class DataAssociation

Defines the type of data association used by the TrackingAlgorithm to associate each cluster to one tracker.

Values:

enumerator NEAREST

Data association is based on the nearest tracker.

enumerator IOU

Data association is based the largest IOU (Intersection Over Union) area between the tracker and the cluster. If no IOU association is available but the distance between the tracker and the cluster is smaller than iou_max_dist_, then the cluster is associated using the nearest criterion. The IOU criterion has the priority over the distance criterion.

enum class Initializer

Defines the type of initializer used by the TrackingAlgorithm to initialize new trackers and make bounding box proposals from input cluster/events. Only one method is implemented for now.

Values:

enumerator SIMPLE

enum class MotionModel

Defines the type of motion model used by the TrackingAlgorithm to predict the positions of trackers.

Values:

enumerator SIMPLE

Motion model that assumes that the velocity is constant.

enumerator INSTANT

Motion model that takes the last measured velocity.

enumerator SMOOTH

Motion model that models the velocity as a smooth evolving quantity parameterized by smooth_mm_alpha_vel_.

enumerator KALMAN

Kalman motion model.

enum class KalmanModel

Defines the motion model used in MotionModel::KALMAN.

Values:

enumerator CONSTANT_VELOCITY

Motion model with a constant velocity.

enumerator CONSTANT_ACCELERATION

Motion model with a constant acceleration.

enum class KalmanPolicy

Defines the policy to determine measurement noise from data in MotionModel::KALMAN.

Values:

enumerator ADAPTIVE_NOISE

Policy that adapts the measurement noise based on the number of events associated to the tracker.

enumerator MEASUREMENT_TRUST

Policy that keeps the noise model as defined in the provided parameters.

enum class Tracker

Defines the type of the tracker used by the TrackingAlgorithm.

Values:

enumerator ELLIPSE

Tracker is based on the event by event update of the tracker’s pose and shape. The tracker is represented as a Gaussian, and the update is performed by weighting each event contribution with EllipseUpdateFunction and updating the tracker’s pose/size using EllipseUpdateMethod.

enumerator CLUSTERKF

Kalman tracker that uses the result of clustering as an observation state to predict the current state of the tracker. It considers the cluster’s barycenter and size as measurements, and it estimates the tracker’s position, velocity and size.

enum class EllipseUpdateFunction

Update function used in Tracker::ELLIPSE.

Values:

enumerator UNIFORM

Function outputs one if the radius is smaller than the parameter, and zero otherwise.

enumerator GAUSSIAN

Function computes a Gaussian with mean equal to 1 and variance given in input.

enumerator SIGNED_GAUSSIAN

Function computes a Gaussian multiplied by the radius minus the mean. This changes the sign of the function inside the radius and gives a negative update.

enumerator TRUNCATED_GAUSSIAN

Function computes a Gaussian if the radius is larger than the mean, otherwise returns zero.

enum class EllipseUpdateMethod

Method used in Tracker::ELLIPSE to update the tracker.

Values:

enumerator PER_EVENT

Method updates the tracker at each event.

enumerator ELLIPSE_FITTING

Method fits an ellipse on the associated cluster events and then updates the tracker’s position event by event and the shape using the covariance of the associated cluster

enumerator GAUSSIAN_FITTING

Method computes the mean and the covariance of the associated cluster events and then updates the tracker’s position event by event and the shape using the covariance of the associated cluster

enumerator ELLIPSE_FITTING_FULL

Method fits an ellipse on the associated cluster events and then updates the tracker using a convex combination on the mean and covariance of the cluster without weight

enumerator GAUSSIAN_FITTING_FULL

Method fits a Gaussian on the associated cluster events and then updates the tracker using a convex combination on the mean and covariance of the cluster without weight

Public Members

bool print_timings_ = false

If enabled, displays a profiling summary.

ClusterMaker cluster_maker_ = ClusterMaker::SIMPLE_GRID

Type of cluster maker to use.

int cell_width_ = 10

Cell’s width (in pixels) in ClusterMaker::SIMPLE_GRID.

int cell_height_ = 10

Cell’s height (in pixels) in ClusterMaker::SIMPLE_GRID.

timestamp cell_deltat_ = 1000

Time-slice (in us) used to decide if a cell is active in ClusterMaker::SIMPLE_GRID.

int activation_threshold_ = 5

Minimum number of events needed to activate the cell in ClusterMaker::SIMPLE_GRID.

float medoid_shift_spatial_dist_ = 5

Maximum spatial distance (using Manhattan distance) for two events to be in the same cluster in ClusterMaker::MEDOID_SHIFT.

timestamp medoid_shift_temporal_dist_ = 10000

Maximum temporal distance for two events to be in the same cluster in ClusterMaker::MEDOID_SHIFT.

int medoid_shift_min_size_ = 2

Minimum width and height for a cluster to be considered valid in ClusterMaker::MEDOID_SHIFT and given to the tracking engine.

DataAssociation data_association_ = DataAssociation::IOU

Type of data association to use.

double max_dist_ = 150.

Maximum distance between the cluster’s centroid and the tracker’s position (in pixels) used in DataAssociation::NEAREST.

double iou_max_dist_ = 150.

Maximum distance between the cluster’s centroid and the tracker’s position (in pixels) used in DataAssociation::IOU. The distance criterion is used only if IOU association is not available.

Initializer initializer_ = Initializer::SIMPLE

Type of tracker initializer to use.

MotionModel motion_model_ = MotionModel::SIMPLE

Type of motion model to use.

double smooth_mm_alpha_vel_ = 0.001

Smoothing parameter used in MotionModel::SMOOTH.

double kalman_motion_model_pos_trans_std_ = 0.0001

Standard deviation of the transition noise for the tracker’s position in MotionModel::KALMAN.

double kalman_motion_model_vel_trans_std_ = 0.05

Standard deviation of the transition noise for the tracker’s velocity in MotionModel::KALMAN.

double kalman_motion_model_acc_trans_std_ = 1e-9

Standard deviation of the transition noise for the tracker’s acceleration in MotionModel::KALMAN.

double kalman_motion_model_pos_obs_std_ = 100

Standard deviation of the observation noise for the tracker’s position in MotionModel::KALMAN.

double kalman_motion_model_init_factor_ = 1e12

Factor to multiply to noise variance at initialization in MotionModel::KALMAN.

double kalman_motion_model_avg_events_per_pixel_ = 1

Expected average events per pixel rate in events/us in MotionModel::KALMAN.

timestamp kalman_motion_model_min_dt_ = 1000

Minimal timestep at which Kalman filter is computed in MotionModel::KALMAN.

KalmanPolicy kalman_motion_model_policy_ = KalmanPolicy::ADAPTIVE_NOISE

Policy used in MotionModel::KALMAN.

KalmanModel kalman_motion_model_motion_model_ = KalmanModel::CONSTANT_VELOCITY

Motion model used in MotionModel::KALMAN.

Tracker tracker_ = Tracker::CLUSTERKF

Type of tracker to use.

double alpha_pos_ = 0.1

Update rate for the tracker position in Tracker::ELLIPSE.

double alpha_shape_ = 0.04

Update rate for the tracker shape in Tracker::ELLIPSE.

EllipseUpdateFunction update_function_ = EllipseUpdateFunction::GAUSSIAN

Update function used in Tracker::ELLIPSE

EllipseUpdateMethod update_method_ = EllipseUpdateMethod::GAUSSIAN_FITTING

Update method used in Tracker::ELLIPSE

bool decompose_covariance_ = false

Flag specifying if to update the covariance decomposing it in eigenvalues and eigenvectors in Tracker::ELLIPSE

double cluster_kf_pos_var_ = 1200000000

Variance of the tracker’s position in Tracker::CLUSTERKF.

double cluster_kf_vel_var_ = 32000

Variance of the tracker’s velocity in Tracker::CLUSTERKF.

double cluster_kf_acc_var_ = 0.8

Variance of the tracker’s acceleration in Tracker::CLUSTERKF.

double cluster_kf_size_var_ = 200000

Variance of the tracker’s size in Tracker::CLUSTERKF.

double cluster_kf_vel_size_var_ = 2

Variance of the tracker’s size change rate in Tracker::CLUSTERKF

double cluster_kf_pos_obs_var_ = 200

Variance of the observed position in Tracker::CLUSTERKF.

double cluster_kf_size_obs_var_ = 1e3

Variance of the observed size in Tracker::CLUSTERKF.

timestamp delta_ts_for_prediction_ = 5000

Delta ts (in us) for the position prediction.

timestamp ts_to_stop_ = 100000

Time (in us) to wait without having updated a tracker before deleting it.

std::filesystem::path forbidden_area_file_ = ""

File indicating the forbidden area.