Detailed Description

Defines ObjectTracker module for object tracking.

Data Structures
struct	dwObjectFeatureTrackerParams

struct	dwObjectTrackerParams

Typedefs
typedef struct dwObjectTrackerObject *	dwObjectTrackerHandle_t
	Handle to an object tracker. More...

Functions
DW_API_PUBLIC dwStatus	dwObjectTracker_boxTrackHost (dwObject trackedDetections, size_t numTrackedDetections, const dwObject *previousDetections, size_t numPreviousDetections, uint32_t classIdx, dwObjectTrackerHandle_t obj)
	Tracks objects in previousDetections based on the tracked features by previously called dwObjectTracker_featureTrackDeviceAsync. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_boxTrackHostExternalFeatures (dwObject trackedDetections, size_t numTrackedDetections, const dwObject previousDetections, size_t numPreviousDetections, const dwFeatureListPointers featureList, uint32_t maxFeatureCount, uint32_t historyCapacity, uint32_t currentTimeIdx, uint32_t classIdx, dwObjectTrackerHandle_t obj)
	This function enables tracking of objects using already extracted features from the previous and current frame. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_featureTrackDeviceAsync (const dwImageCUDA *image, dwObjectTrackerHandle_t obj)
	Tracks the features from a previously given image to the current image. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_getCUDAStream (cudaStream_t *stream, dwObjectTrackerHandle_t obj)
	Gets the CUDA stream used. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_initDefaultParams (dwObjectFeatureTrackerParams featureTrackerParams, dwObjectTrackerParams objectTrackerParamsArray, uint32_t numClasses)
	Initializes ObjectTracker parameters with default values. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_initialize (dwObjectTrackerHandle_t obj, dwContextHandle_t ctx, const dwImageProperties imageProperties, const dwObjectFeatureTrackerParams featureTrackerParams, const dwObjectTrackerParams objectTrackerParamsArray, uint32_t numClasses)
	Initializes the ObjectTracker module. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_release (dwObjectTrackerHandle_t *obj)
	Releases the ObjectTracker module. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_reset (dwObjectTrackerHandle_t obj)
	Resets the ObjectTracker. More...

DW_API_PUBLIC dwStatus	dwObjectTracker_setCUDAStream (cudaStream_t stream, dwObjectTrackerHandle_t obj)
	Sets the CUDA stream used. More...

Data Structure Documentation

◆ dwObjectFeatureTrackerParams

struct dwObjectFeatureTrackerParams

Data Fields
float32_t	detectorScoreThreshold	Higher thresholds detect less features. Default is 0.3.
uint32_t	historyCapacity	Capacity of feature history circular buffer.
uint32_t	iterationsLK	Upper bound on number of Lucas-Kanade iterations per level. Default is 10.
dwRect	maskROI	Region of interest where the objects are tracked. Default is full image.
uint32_t	maxFeatureCount	Upper bound on number of features handled.
uint32_t	pyramidLevelCount	Number of levels in the pyramid. Default is 3.
uint32_t	windowSizeLK	Window size used in the Lucas-Kanade tracker. Supported sizes are 6,8,10,12,14. Default is 8.

◆ dwObjectTrackerParams

struct dwObjectTrackerParams

Data Fields
float32_t	confRateDetect	Rate at which to combine confidence values of new boxes to existing tracked boxes when a new box is found to match an existing box, i.e., conf = conf + conf_new * rate. This also applies to the initial confidence value for a new box, i.e., conf = conf_new * rate. Default is 1.
float32_t	confRateTrackMax	Max rate at which confidence values of tracked boxes changes from frame to frame. When there is no corresponding detection, conf = conf - actualConfRate. Actual decaying rate is determined between 'confRateTrackMin' and 'confRateTrackMax' according to the number of feature points used to track the object. i.e., actualConfRate = confRateTrackMin + (confRateTrackMax - confRateTrackMin) * (1.f - sqrt(featureCount / maxFeatureCount))
float32_t	confRateTrackMin	Min rate at which confidence values of tracked boxes changes from frame to frame.
float32_t	confThreshConfirm	Threshold on confidence below which tracker will not confirmed to be used yet.
float32_t	confThreshDiscard	Threshold on confidence below which tracker no longer tracks the box location. Default is 0.
dwBool	enablePriorityTracking	Priority of the objects can be used to control association between 2D features and bounding boxes in case of overlap. With priority tracking enabled, the object with the lower index gets the feature.
float32_t	maxBoxImageScale	Maximum box scale in the image to track. It multiplies with image width and height to get the maximum box size. Default is 0.5.
uint32_t	maxFeatureCountPerBox	Maximum features to track for each 2D bounding box. Default is 500.
uint32_t	maxNumObjects	Maximum number of objects to track. This information is used to allocate resources at initialization time. Default is 100.
float32_t	minBoxImageScale	Minimum box scale in the image to track. It multiplies with image width and height to get the minimum box size. Default is 0.005.

Typedef Documentation

◆ dwObjectTrackerHandle_t

typedef struct dwObjectTrackerObject* dwObjectTrackerHandle_t

Handle to an object tracker.

Definition at line 62 of file Tracker.h.

Function Documentation

◆ dwObjectTracker_boxTrackHost()

DW_API_PUBLIC dwStatus dwObjectTracker_boxTrackHost	(	dwObject *	trackedDetections,
		size_t *	numTrackedDetections,
		const dwObject *	previousDetections,
		size_t	numPreviousDetections,
		uint32_t	classIdx,
		dwObjectTrackerHandle_t	obj
	)

Tracks objects in previousDetections based on the tracked features by previously called dwObjectTracker_featureTrackDeviceAsync.

This function synchronizes host with the CUDA stream.

Parameters

[out]	trackedDetections	List of tracked objects.
[out]	numTrackedDetections	Number of tracked objects.
[in]	previousDetections	List of input objects to be tracked.
[in]	numPreviousDetections	Number of input objects.
[in]	classIdx	Class index from which to use the object tracker parameters.
[in]	obj	Specifies the ObjectTracker handle.

Returns: DW_SUCCESS, DW_INVALID_HANDLE, DW_BAD_CAST, DW_INVALID_ARGUMENT

Note: trackedDetections must be allocated with maxNumObjects given at initialization.

◆ dwObjectTracker_boxTrackHostExternalFeatures()

DW_API_PUBLIC dwStatus dwObjectTracker_boxTrackHostExternalFeatures	(	dwObject *	trackedDetections,
		size_t *	numTrackedDetections,
		const dwObject *	previousDetections,
		size_t	numPreviousDetections,
		const dwFeatureListPointers *	featureList,
		uint32_t	maxFeatureCount,
		uint32_t	historyCapacity,
		uint32_t	currentTimeIdx,
		uint32_t	classIdx,
		dwObjectTrackerHandle_t	obj
	)

This function enables tracking of objects using already extracted features from the previous and current frame.

This function synchronizes host with the CUDA stream.

Parameters

[out]	trackedDetections	List of tracked objects.
[out]	numTrackedDetections	Number of tracked objects.
[in]	previousDetections	List of input objects to be tracked.
[in]	numPreviousDetections	Number of input objects.
[in]	featureList	Feature list in host.
[in]	maxFeatureCount	Maximum number of features that featureList can store.
[in]	historyCapacity	Capacity of featureList circular buffer.
[in]	currentTimeIdx	Row index of the current frame in featureList circular buffer.
[in]	classIdx	Class index from which to use the object tracker parameters.
[in]	obj	Specifies the ObjectTracker handle.

Returns: DW_SUCCESS, DW_INVALID_HANDLE, DW_BAD_CAST, DW_INVALID_ARGUMENT

Note: trackedDetections must be allocated with maxNumObjects given at initialization.; previousFeatures and currentFeatures are expected to be in host memory.

◆ dwObjectTracker_featureTrackDeviceAsync()

DW_API_PUBLIC dwStatus dwObjectTracker_featureTrackDeviceAsync	(	const dwImageCUDA *	image,
		dwObjectTrackerHandle_t	obj
	)

Tracks the features from a previously given image to the current image.

This function must be called before dwObjectTracker_boxTrackHost. The computation takes place asynchronously on the device (GPU).

Parameters

[in]	image	CUDA image where the objects are tracked.
[in]	obj	Specifies the ObjectTracker handle.

Returns: DW_SUCCESS, DW_INVALID_HANDLE, DW_BAD_CAST, DW_INVALID_ARGUMENT

◆ dwObjectTracker_getCUDAStream()

DW_API_PUBLIC dwStatus dwObjectTracker_getCUDAStream	(	cudaStream_t *	stream,
		dwObjectTrackerHandle_t	obj
	)

Gets the CUDA stream used.

Parameters

[out]	stream	The CUDA stream used.
[in]	obj	A pointer to the ObjectTracker handle that is updated.

Returns: DW_INVALID_ARGUMENT, DW_SUCCESS

◆ dwObjectTracker_initDefaultParams()

DW_API_PUBLIC dwStatus dwObjectTracker_initDefaultParams	(	dwObjectFeatureTrackerParams *	featureTrackerParams,
		dwObjectTrackerParams *	objectTrackerParamsArray,
		uint32_t	numClasses
	)

Initializes ObjectTracker parameters with default values.

Parameters

[out]	featureTrackerParams	Feature tracking parameters.
[out]	objectTrackerParamsArray	Object tracking parameters for each class.
[in]	numClasses	Number of classes. Maximum allowed is DW_OBJECT_MAX_CLASSES.

Returns: DW_SUCCESS, DW_INVALID_ARGUMENT

◆ dwObjectTracker_initialize()

DW_API_PUBLIC dwStatus dwObjectTracker_initialize	(	dwObjectTrackerHandle_t *	obj,
		dwContextHandle_t	ctx,
		const dwImageProperties *	imageProperties,
		const dwObjectFeatureTrackerParams *	featureTrackerParams,
		const dwObjectTrackerParams *	objectTrackerParamsArray,
		uint32_t	numClasses
	)

Initializes the ObjectTracker module.

Parameters

[out]	obj	A pointer to ObjectTracker handle that is initialized from parameters.
[in]	ctx	Specifies the handle to the context.
[in]	imageProperties	Image properties to set up the tracker for.
[in]	featureTrackerParams	Feature tracker parameters.
[in]	objectTrackerParamsArray	Object tracker parameters for each object class.
[in]	numClasses	Number of object classes.

Returns: DW_SUCCESS, DW_INVALID_HANDLE, DW_BAD_CAST, DW_INVALID_ARGUMENT

Note: Expected pxlType for the image are: DW_TYPE_UINT8, DW_TYPE_UINT16, DW_TYPE_FP16

◆ dwObjectTracker_release()

DW_API_PUBLIC dwStatus dwObjectTracker_release ( dwObjectTrackerHandle_t * obj )

Releases the ObjectTracker module.

Parameters

[in] obj A pointer to the object handle to release.

Returns: DW_SUCCESS, DW_INVALID_HANDLE, DW_BAD_CAST

Note: This method renders the handle unusable.

◆ dwObjectTracker_reset()

DW_API_PUBLIC dwStatus dwObjectTracker_reset ( dwObjectTrackerHandle_t obj )

Resets the ObjectTracker.

Parameters

[in] obj Handle to reset.

Returns: DW_SUCCESS, DW_INVALID_HANDLE, DW_BAD_CAST

◆ dwObjectTracker_setCUDAStream()

DW_API_PUBLIC dwStatus dwObjectTracker_setCUDAStream	(	cudaStream_t	stream,
		dwObjectTrackerHandle_t	obj
	)

Sets the CUDA stream used.

Parameters

[in]	stream	The CUDA stream used.
[in]	obj	A pointer to the ObjectTracker handle that is updated.

Returns: DW_INVALID_ARGUMENT, DW_SUCCESS

DriveWorks SDK Reference

| 0.6.67 Release

Detailed Description

Data Structures

Typedefs

Functions

Data Structure Documentation

◆ dwObjectFeatureTrackerParams

◆ dwObjectTrackerParams

Typedef Documentation

◆ dwObjectTrackerHandle_t

Function Documentation

◆ dwObjectTracker_boxTrackHost()

◆ dwObjectTracker_boxTrackHostExternalFeatures()

◆ dwObjectTracker_featureTrackDeviceAsync()

◆ dwObjectTracker_getCUDAStream()

◆ dwObjectTracker_initDefaultParams()

◆ dwObjectTracker_initialize()

◆ dwObjectTracker_release()

◆ dwObjectTracker_reset()

◆ dwObjectTracker_setCUDAStream()