Skip to content

Latest commit

 

History

History
391 lines (303 loc) · 17.1 KB

api-tracker.md

File metadata and controls

391 lines (303 loc) · 17.1 KB
Raw

Multi-Object Tracker API Reference

The DeepStream Services Library (DSL) supports Nvidia's four reference low-level trackers (Note: the below bullets are copied from the Nvidia DeepStream Gst-nvtracker plugin-guide).

  • IOU Tracker: The Intersection-Over-Union (IOU) tracker uses the IOU values among the detector’s bounding boxes between the two consecutive frames to perform the association between them or assign a new target ID if no match found. This tracker includes a logic to handle false positives and false negatives from the object detector; however, this can be considered as the bare-minimum object tracker, which may serve as a baseline only.
  • NvSORT: The NvSORT tracker is the NVIDIA®-enhanced Simple Online and Realtime Tracking (SORT) algorithm. Instead of a simple bipartite matching algorithm, NvSORT uses a cascaded data association based on bounding box (bbox) proximity for associating bboxes over consecutive frames and applies a Kalman filter to update the target states. It is computationally efficient since it does not involve any pixel data processing.
  • DeepSORT: The DeepSORT tracker is a re-implementation of the official DeepSORT tracker, which uses the deep cosine metric learning with a Re-ID neural network. This implementation allows users to use any Re-ID network as long as it is supported by NVIDIA’s TensorRT™ framework.
  • NvDCF: The NvDCF tracker is an NVIDIA®-adapted Discriminative Correlation Filter (DCF) tracker that uses a correlation filter-based online discriminative learning algorithm for visual object tracking capability, while using a data association algorithm and a state estimator for multi-object tracking.

The four reference implementations are provided in a single low-level library libnvds_nvmultiobjecttracker.so which is specified as the default library for the DSL Tracker to use as defined in the Makefile. The default library path can be updated -- by updating the Makefile or by calling dsl_tracker_lib_file_set -- to reference a custom library that implements the NvDsTracker API.

Important! The DeepSORT tracker requires additional installation and setup steps which can be found in the README file located under

/opt/nvidia/deepstream/deepstream/sources/tracker_DeepSORT

Construction and Destruction

A Tracker component is created by calling dsl_tracker_new with a type specific configuration file.

Important! NVIDIA® provides reference configuration files for the four Tracker implementations under

/opt/nvidia/deepstream/deepstream/samples/configs/deepstream-app/

Tracker components are deleted by calling dsl_component_delete, dsl_component_delete_many, or dsl_component_delete_all. Calling a delete service on a Tracker in-use by a Pipeline will fail.

Adding and Removing

The relationship between Pipelines/Branches and Trackers is one-to-one. Once added to a Pipeline or Branch, a Tracker must be removed before it can be used with another. A Tracker is added to a Pipeline by calling dsl_pipeline_component_add or dsl_pipeline_component_add_many and removed with dsl_pipeline_component_remove, dsl_pipeline_component_remove_many, or dsl_pipeline_component_remove_all.

A similar set of Services are used when adding/removing a to/from a branch: dsl_branch_component_add, dsl_branch_component_add_many, dsl_branch_component_remove, dsl_branch_component_remove_many, and dsl_branch_component_remove_all.

Pipelines with a Tracker component require a Primary GIE/TIS component in order to Play.

Adding/Removing Pad-Probe-handlers

Multiple sink and/or source Pad-Probe Handlers can be added to a Tracker by calling dsl_tracker_pph_add and removed with dsl_tracker_pph_remove.

Tracker API

Constructors:

Methods:


Return Values

The following return codes are used specifically by the Tracker API

#define DSL_RESULT_TRACKER_RESULT                                   0x00030000
#define DSL_RESULT_TRACKER_NAME_NOT_UNIQUE                          0x00030001
#define DSL_RESULT_TRACKER_NAME_NOT_FOUND                           0x00030002
#define DSL_RESULT_TRACKER_NAME_BAD_FORMAT                          0x00030003
#define DSL_RESULT_TRACKER_THREW_EXCEPTION                          0x00030004
#define DSL_RESULT_TRACKER_CONFIG_FILE_NOT_FOUND                    0x00030005
#define DSL_RESULT_TRACKER_IS_IN_USE                                0x00030006
#define DSL_RESULT_TRACKER_SET_FAILED                               0x00030007
#define DSL_RESULT_TRACKER_HANDLER_ADD_FAILED                       0x00030008
#define DSL_RESULT_TRACKER_HANDLER_REMOVE_FAILED                    0x00030009

Constructors

dsl_tracker_new

DslReturnType dsl_tracker_new(const wchar_t* name, const wchar_t* config_file,
    uint width, uint height);

This service creates a uniquely named Tracker component using the default NVIDIA NvMultiObjectTracker low-level library. Construction will fail if the name is currently in use. The config_file parameter is optional. If not specified, the low-level library will proceed with default IOU settings.

Note a custom implementation of the NvDsTracker API can be used by setting the Tracker's low-level library by calling dsl_tracker_lib_file_set.

Parameters

  • name - [in] unique name for the Tracker to create.
  • config_file - [in] relative or absolute pathspec to a valid config text file. Set to NULL or empty string to omit.
  • width - [in] Frame width at which the tracker is to operate, in pixels.
  • height - [in] Frame height at which the tracker is to operate, in pixels.

Returns

  • DSL_RESULT_SUCCESS on successful creation. One of the Return Values defined above on failure

Python Example

retval = dsl_tracker_new('my-iou-tracker', './test/configs/iou_config.txt', 640, 384)

Methods

dsl_tracker_lib_file_get

DslReturnType dsl_tracker_lib_file_get(const wchar_t* name,
    const wchar_t** lib_file);

This service returns the absolute path to the low-level library in use by the named Tracker.

Important the default path to low-level library is defined in the Makefile as $(LIB_INSTALL_DIR)/libnvds_nvmultiobjecttracker.so

Parameters

  • name - [in] unique name of the Tracker to query.
  • lib_file - [out] absolute pathspec to the low-level library in use.

Returns

  • DSL_RESULT_SUCCESS on successful query. One of the Return Values defined above on failure

Python Example

retval, lib_file = dsl_tracker_lib_file_get('my-tracker')

dsl_tracker_lib_file_set

DslReturnType dsl_tracker_lib_file_set(const wchar_t* name,
    const wchar_t* lib_file);

This service updates the named Tracker with a new low-level library to use.

Parameters

  • name - [in] unique name for the Tracker to update.
  • lib_file - [in] absolute or relative pathspec to the new low-level library to use.

Returns

  • DSL_RESULT_SUCCESS on successful update. One of the Return Values defined above on failure

Python Example

retval = dsl_tracker_lib_file_set('my-tracker', path_to_ll_lib)

dsl_tracker_config_file_get

DslReturnType dsl_tracker_config_file_get(const wchar_t* name,
    const wchar_t** config_file);

This service returns the absolute path to the (optional) Tracker Config File in use by the named Tracker. This service returns an empty string if the configuration file was omitted on construction, or removed by calling dsl_tracker_config_file_set with a NULL pointer.

Parameters

  • name - [in] unique name of the Tracker to query.
  • config_file - [out] absolute pathspec to the config text file in use.

Returns

  • DSL_RESULT_SUCCESS on successful query. One of the Return Values defined above on failure

Python Example

retval, config_file = dsl_tracker_config_file_get('my-tracker')

dsl_tracker_config_file_set

DslReturnType dsl_tracker_iou_config_file_set(const wchar_t* name,
    const wchar_t* config_file);

This service updates the named Tracker with a new config file to use.

Parameters

  • name - [in] unique name of the Tracker to update.
  • config_file - [in] absolute pathspec to the config text file in use. Set config_file to NULL to clear the optional configuration file setting.

Returns

  • DSL_RESULT_SUCCESS on successful update. One of the Return Values defined above on failure

Python Example

retval = dsl_tracker_config_file_set('my-tracker', './test/configs/iou_config.txt')

dsl_tracker_dimensions_get

DslReturnType dsl_tracker_dimensions_get(const wchar_t* name, uint* width, uint* height);

This service returns the operational dimensions in use by the named Tracker.

Parameters

  • name - [in] unique name of the Tracker to query.
  • width - [out] Current frame width at which the tracker is to operate, in pixels.
  • height - [out] Current frame height at which the tracker is to operate, in pixels.

Returns

  • DSL_RESULT_SUCCESS on successful query. One of the Return Values defined above on failure

Python Example

retval, width, height = dsl_tracker_dimensions_get('my-tracker')

dsl_tracker_dimensions_set

DslReturnType dsl_tracker_dimensions_set(const wchar_t* name, uint max_width, uint max_height);

This Service sets the operational dimensions for the name Tracker.

Parameters

  • name - [in] unique name of the Tracker to update.
  • width - [in] Frame width at which the tracker is to operate, in pixels.
  • height - [in] Frame height at which the tracker is to operate, in pixels.

Returns

  • DSL_RESULT_SUCCESS on successful update. One of the Return Values defined above on failure

Python Example

retval = dsl_tracker_dimensions_set('my-tracker', 640, 368)

dsl_tracker_tensor_meta_settings_get

DslReturnType dsl_tracker_tensor_meta_settings_get(const wchar_t* name, 
    boolean* input_enabled, const wchar_t** track_on_gie);

This service gets the current current tensor-meta settings for the named Tracker.

Parameters

  • name - [in] unique name of the Tracker to query.
  • input_enabled - [out] if true, Tracker uses the tensor-meta from the Preprocessor if available and the PGIE identified by track_on_gie.
  • track_on_gie - [out] name of the PGIE to track on if input_enabled.

Returns

  • DSL_RESULT_SUCCESS on successful query. One of the Return Values defined above on failure

Python Example

retval, input_enabled, track_on_gie = dsl_tracker_tensor_meta_settings_get('my-tracker')

dsl_tracker_tensor_meta_settings_set

DslReturnType dsl_tracker_tensor_meta_settings_set(const wchar_t* name, 
    boolean input_enabled, const wchar_t* track_on_gie);

This service sets the tensor-meta settings for the named Tracker.

Parameters

  • name - [in] unique name of the Tracker to update.
  • input_enabled - [in] if true, Tracker uses the tensor-meta from the Preprocessor if available and the PGIE identified by track_on_gie.
  • track_on_gie - [in] name of the PGIE to track on if input_enabled.

Returns

  • DSL_RESULT_SUCCESS on successful update. One of the Return Values defined above on failure

Python Example

retval = dsl_tracker_tensor_meta_settings_set('my-tracker', True, 'my-pgie')

dsl_tracker_id_display_enabled_get

DslReturnType dsl_tracker_id_display_enabled_get(const wchar_t* name, 
    boolean* enabled);

This service gets the current "tracker-id-display-enabled" setting for the named Tracker.

Parameters

  • name - [in] unique name of the Tracker to query.
  • enabled - [out] if true, tracking-ids will included in object labels. Default = True.

Returns

  • DSL_RESULT_SUCCESS on successful query. One of the Return Values defined above on failure

Python Example

retval, enabled = dsl_tracker_id_display_enabled_get('my-tracker')

dsl_tracker_id_display_enabled_set

DslReturnType dsl_tracker_id_display_enabled_set(const wchar_t* name, 
    boolean enabled);

This service sets the "tracker-id-display-enabled" setting for the named Tracker object.

Parameters

  • name - [in] unique name of the Tracker to update.
  • enabled - [in] set to true to included tracking-ids in object labels, false to exclude.

Returns

  • DSL_RESULT_SUCCESS on successful update. One of the Return Values defined above on failure

Python Example

retval = dsl_tracker_id_display_enabled_set('my-tracker', False)

dsl_tracker_pph_add

DslReturnType dsl_tracker_pph_add(const wchar_t* name, const wchar_t* handler, uint pad);

This service adds a Pad Probe Handler to either the Sink or Source pad of the named Tracker.

Parameters

  • name - [in] unique name of the Tracker to update.
  • handler - [in] unique name of Pad Probe Handler to add
  • pad - [in] to which of the two pads to add the handler: DSL_PAD_SIK or DSL_PAD SRC

Returns

  • DSL_RESULT_SUCCESS on successful add. One of the Return Values defined above on failure.

Python Example

retval = dsl_tracker_pph_add('my-tracker', 'my-pph-handler', `DSL_PAD_SINK`)

dsl_tracker_pph_remove

DslReturnType dsl_tracker_pph_remove(const wchar_t* name, const wchar_t* handler, uint pad);

This service removes a Pad Probe Handler from either the Sink or Source pad of the named Tracker. The service will fail if the named handler is not owned by the Tracker

Parameters

  • name - [in] unique name of the Tracker to update.
  • handler - [in] unique name of Pad Probe Handler to remove
  • pad - [in] to which of the two pads to remove the handler from: DSL_PAD_SIK or DSL_PAD SRC

Returns

  • DSL_RESULT_SUCCESS on successful removal. One of the Return Values defined above on failure.

Python Example

retval = dsl_tracker_pph_remove('my-tracker', 'my-pph-handler', `DSL_PAD_SINK`)

API Reference