Functions
Eigen::VectorXd	convertBboxToZ (const NvMotDataTypes::Rect2f &bbox, const bool useAspc=false)
	Convert a bounding box to an estimator measurement vector. More...

void	convexHull (const std::vector< Eigen::Vector2d > &points, std::vector< Eigen::Vector2d > &hull)
	Compute the convex hull of double-precision points. More...

void	convexHull (const std::vector< Eigen::Vector2f > &points, std::vector< Eigen::Vector2f > &hull)
	Compute the convex hull of single-precision points. More...

NvMotDataTypes::Rect2f	boundingRect (const std::vector< Eigen::Vector2d > &points)
	Compute the axis-aligned bounding rectangle of a point set. More...

NvMotDataTypes::Rect2f	boundingRect (float *mask, const int width, const int height)
	Compute the bounding rectangle of nonzero mask values. More...

NvMotDataTypes::Rect2f	constructRect2f (const float minX=0, const float minY=0, const float width=0, const float height=0)
	Construct an axis-aligned rectangle from its origin and size. More...

NvMotDataTypes::Rect2f	displaceBbox (const NvMotDataTypes::Rect2f &bbox, const float displacement_x, const float displacement_y)
	Translate a bounding box by horizontal and vertical displacements. More...

Eigen::Vector2f	getBboxCenterPt (const NvMotDataTypes::Rect2f &bbox)
	Compute the center point of a bounding box. More...

const std::string	getEigenRectString (const NvMotDataTypes::Rect2f &bbox)
	Format an Eigen bounding box as a readable string. More...

bool	isBboxEqual (const NvMotDataTypes::Rect2f &bbox1, const NvMotDataTypes::Rect2f &bbox2)
	Compare two bounding boxes for exact equality. More...

int	modul (int a, int b)
	Compute a nonnegative modulo result for positive divisors. More...

NvMotDataTypes::Rect2f	getBoxWithAspectRatio (const NvMotDataTypes::Rect2f &bbox, const float aspectRatio)
	Adjust the width of the bbox according to the aspect ratio specified. More...

NvMotDataTypes::Rect2f	convertXToBbox (const Eigen::VectorXd &x, const bool useAspc=false)
	Convert an estimator state vector to a bounding box. More...

NvMotDataTypes::RowMajorMatrixXf	getHanningWindow (const int width, const int height, float offsetFactor_y=0.f)
	Generate a separable two-dimensional Hann window. More...

float	computeSizeScore (const NvMotDataTypes::Rect2f &currBox, const NvMotDataTypes::Rect2f &prevBox)
	Compute a symmetric area-ratio score between two bounding boxes. More...

Eigen::MatrixXf	circshift (const Eigen::MatrixXf &matrix, int dx, int dy)
	Circularly shift a matrix along both axes. More...

Eigen::MatrixXf	get2dGaussianPdf (const float sigma, const int w, const int h, const bool wrap_kernel=true)
	Create a two-dimensional Gaussian response map. More...

const NvMotDataTypes::Rect2f	unclipBbox (const NvMotDataTypes::Rect2f tBbox_new, const NvMotDataTypes::Rect2f srcImageBox, const NvMotDataTypes::Rect2f tBbox_prev)
	Recover bounding-box extent clipped by an image border using the previous box size. More...

const NvMotDataTypes::Rect2f	computeSearchRegion (const NvMotDataTypes::Rect2f &bbox_new, const float paddingScale)
	Expand a bounding box into a square search region. More...

float	computeIOU (const NvMotDataTypes::Rect2f currBox, const NvMotDataTypes::Rect2f prevBox)
	Compute intersection over union for two bounding boxes. More...

float	computeVisibility (const NvMotDataTypes::Rect2f &objBbox, const NvMotDataTypes::Rect2f &imgBbox)
	Compute the fraction of an object bounding box visible inside an image box. More...

int	orientation (const Eigen::Vector2d &p, const Eigen::Vector2d &q, const Eigen::Vector2d &r)
	Classify the orientation of three double-precision points. More...

int	orientation (const Eigen::Vector2f &p, const Eigen::Vector2f &q, const Eigen::Vector2f &r)
	Classify the orientation of three single-precision points. More...

Function Documentation

◆ boundingRect() [1/2]

NvMotDataTypes::Rect2f CV_FUNCS::boundingRect ( const std::vector< Eigen::Vector2d > & points )

Compute the axis-aligned bounding rectangle of a point set.

Parameters

[in] points Input points.

◆ boundingRect() [2/2]

NvMotDataTypes::Rect2f CV_FUNCS::boundingRect	(	float *	mask,
		const int	width,
		const int	height
	)

Compute the bounding rectangle of nonzero mask values.

Parameters

[in]	mask	Row-major mask values.
[in]	width	Mask width.
[in]	height	Mask height.

◆ circshift()

Eigen::MatrixXf CV_FUNCS::circshift	(	const Eigen::MatrixXf &	matrix,
		int	dx,
		int	dy
	)

inline

Circularly shift a matrix along both axes.

Parameters

[in]	matrix	Matrix to shift.
[in]	dx	Horizontal shift.
[in]	dy	Vertical shift.

Definition at line 293 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References modul().

Referenced by get2dGaussianPdf(), and TrackerVizOutput::prepareCorrelationResponseImg().

◆ computeIOU()

float CV_FUNCS::computeIOU	(	const NvMotDataTypes::Rect2f	currBox,
		const NvMotDataTypes::Rect2f	prevBox
	)

inline

Compute intersection over union for two bounding boxes.

Parameters

[in]	currBox	Current bounding box.
[in]	prevBox	Previous bounding box.

Definition at line 425 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

◆ computeSearchRegion()

const NvMotDataTypes::Rect2f CV_FUNCS::computeSearchRegion	(	const NvMotDataTypes::Rect2f &	bbox_new,
		const float	paddingScale
	)

inline

Expand a bounding box into a square search region.

Parameters

[in]	bbox_new	Bounding box around which the search region is centered.
[in]	paddingScale	Padding scale applied to the geometric mean box size.

Define search region for the update.

Definition at line 394 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References constructRect2f(), floorf(), and half_float::detail::sqrt().

◆ computeSizeScore()

float CV_FUNCS::computeSizeScore	(	const NvMotDataTypes::Rect2f &	currBox,
		const NvMotDataTypes::Rect2f &	prevBox
	)

inline

Compute a symmetric area-ratio score between two bounding boxes.

Parameters

[in]	currBox	Current bounding box.
[in]	prevBox	Previous bounding box.

Definition at line 270 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

◆ computeVisibility()

float CV_FUNCS::computeVisibility	(	const NvMotDataTypes::Rect2f &	objBbox,
		const NvMotDataTypes::Rect2f &	imgBbox
	)

inline

Compute the fraction of an object bounding box visible inside an image box.

Parameters

[in]	objBbox	Object bounding box.
[in]	imgBbox	Image bounding box.

Definition at line 459 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

◆ constructRect2f()

NvMotDataTypes::Rect2f CV_FUNCS::constructRect2f	(	const float	minX = `0`,
		const float	minY = `0`,
		const float	width = `0`,
		const float	height = `0`
	)

inline

Construct an axis-aligned rectangle from its origin and size.

Parameters

[in]	minX	Left coordinate.
[in]	minY	Top coordinate.
[in]	width	Rectangle width.
[in]	height	Rectangle height.

Definition at line 93 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References max(), min(), and width.

Referenced by computeSearchRegion(), convertXToBbox(), DetectedObjInfo::DetectedObjInfo(), and TrackerVizOutput::drawRecentTrajectory().

◆ convertBboxToZ()

Eigen::VectorXd CV_FUNCS::convertBboxToZ	(	const NvMotDataTypes::Rect2f &	bbox,
		const bool	useAspc = `false`
	)

Convert a bounding box to an estimator measurement vector.

Parameters

[in]	bbox	Bounding box to convert.
[in]	useAspc	Selects aspect ratio instead of width in the measurement vector.

◆ convertXToBbox()

NvMotDataTypes::Rect2f CV_FUNCS::convertXToBbox	(	const Eigen::VectorXd &	x,
		const bool	useAspc = `false`
	)

inline

Convert an estimator state vector to a bounding box.

Parameters

[in]	x	Estimator state vector.
[in]	useAspc	Selects aspect ratio instead of width in the state vector.

Derive the width from the aspect ratio when the state uses one.

Definition at line 217 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References constructRect2f().

Referenced by NvTrackedObject::getPredictedTargetBbox().

◆ convexHull() [1/2]

void CV_FUNCS::convexHull	(	const std::vector< Eigen::Vector2d > &	points,
		std::vector< Eigen::Vector2d > &	hull
	)

Compute the convex hull of double-precision points.

Parameters

[in]	points	Input points.
[out]	hull	Convex-hull vertices in traversal order.

Referenced by TrackerVizOutput::drawProjectedTargetModel().

◆ convexHull() [2/2]

void CV_FUNCS::convexHull	(	const std::vector< Eigen::Vector2f > &	points,
		std::vector< Eigen::Vector2f > &	hull
	)

Compute the convex hull of single-precision points.

Parameters

[in]	points	Input points.
[out]	hull	Convex-hull vertices in traversal order.

◆ displaceBbox()

NvMotDataTypes::Rect2f CV_FUNCS::displaceBbox	(	const NvMotDataTypes::Rect2f &	bbox,
		const float	displacement_x,
		const float	displacement_y
	)

inline

Translate a bounding box by horizontal and vertical displacements.

Parameters

[in]	bbox	Bounding box to translate.
[in]	displacement_x	Horizontal displacement.
[in]	displacement_y	Vertical displacement.

Update the target center from the displacement in correlation response.

Definition at line 109 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References max(), and min().

◆ get2dGaussianPdf()

Eigen::MatrixXf CV_FUNCS::get2dGaussianPdf	(	const float	sigma,
		const int	w,
		const int	h,
		const bool	wrap_kernel = `true`
	)

inline

Create a two-dimensional Gaussian response map.

Parameters

[in]	sigma	Gaussian standard deviation.
[in]	w	Response-map width.
[in]	h	Response-map height.
[in]	wrap_kernel	Circularly shifts the peak to the origin when true.

Calculate for each pixel separately.

Wrap-around with the circular shifting.

Calculate for each pixel separately.

Wrap-around with the circular shifting.

Calculate for each pixel separately.

Wrap-around with the circular shifting.

Definition at line 317 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References circshift(), half_float::exp(), and half_float::pow().

◆ getBboxCenterPt()

Eigen::Vector2f CV_FUNCS::getBboxCenterPt ( const NvMotDataTypes::Rect2f & bbox )

inline

Compute the center point of a bounding box.

Parameters

[in] bbox Bounding box whose center is requested.

Definition at line 128 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References min(), and pydeepstreamdoc::tensor::TensorDoc::size.

Referenced by DetectedObjInfo::DetectedObjInfo().

◆ getBoxWithAspectRatio()

NvMotDataTypes::Rect2f CV_FUNCS::getBoxWithAspectRatio	(	const NvMotDataTypes::Rect2f &	bbox,
		const float	aspectRatio
	)

inline

Adjust the width of the bbox according to the aspect ratio specified.

Parameters

[in]	bbox	Bounding box whose width is adjusted to the given aspect ratio.
[in]	aspectRatio	Width divided by height used to resize the box.

Returns: Adjusted bounding box.

Definition at line 173 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

◆ getEigenRectString()

const std::string CV_FUNCS::getEigenRectString ( const NvMotDataTypes::Rect2f & bbox )

inline

Format an Eigen bounding box as a readable string.

Parameters

[in] bbox Bounding box to format.

Definition at line 140 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

◆ getHanningWindow()

NvMotDataTypes::RowMajorMatrixXf CV_FUNCS::getHanningWindow	(	const int	width,
		const int	height,
		float	offsetFactor_y = `0.f`
	)

inline

Generate a separable two-dimensional Hann window.

Parameters

[in]	width	Window width.
[in]	height	Window height.
[in]	offsetFactor_y	Vertical offset as a fraction of window height.

Definition at line 235 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References half_float::cos(), and width.

◆ isBboxEqual()

bool CV_FUNCS::isBboxEqual	(	const NvMotDataTypes::Rect2f &	bbox1,
		const NvMotDataTypes::Rect2f &	bbox2
	)

inline

Compare two bounding boxes for exact equality.

Parameters

[in]	bbox1	First bounding box.
[in]	bbox2	Second bounding box.

Definition at line 152 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

◆ modul()

int CV_FUNCS::modul	(	int	a,
		int	b
	)

inline

Compute a nonnegative modulo result for positive divisors.

Parameters

[in]	a	Dividend.
[in]	b	Divisor.

Definition at line 162 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

Referenced by circshift().

◆ orientation() [1/2]

int CV_FUNCS::orientation	(	const Eigen::Vector2d &	p,
		const Eigen::Vector2d &	q,
		const Eigen::Vector2d &	r
	)

inline

Classify the orientation of three double-precision points.

Parameters

[in]	p	First point.
[in]	q	Second point.
[in]	r	Third point.

Returns: Zero for collinear points, one for clockwise order, or two for counterclockwise order.

Return the collinear classification.

Classify the turn as clockwise or counterclockwise.

Return the collinear classification.

Classify the turn as clockwise or counterclockwise.

Return the collinear classification.

Classify the turn as clockwise or counterclockwise.

Definition at line 493 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References val.

◆ orientation() [2/2]

int CV_FUNCS::orientation	(	const Eigen::Vector2f &	p,
		const Eigen::Vector2f &	q,
		const Eigen::Vector2f &	r
	)

inline

Classify the orientation of three single-precision points.

Parameters

[in]	p	First point.
[in]	q	Second point.
[in]	r	Third point.

Returns: Zero for collinear points, one for clockwise order, or two for counterclockwise order.

Return the collinear classification.

Classify the turn as clockwise or counterclockwise.

Return the collinear classification.

Classify the turn as clockwise or counterclockwise.

Return the collinear classification.

Classify the turn as clockwise or counterclockwise.

Definition at line 511 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

References val.

◆ unclipBbox()

const NvMotDataTypes::Rect2f CV_FUNCS::unclipBbox	(	const NvMotDataTypes::Rect2f	tBbox_new,
		const NvMotDataTypes::Rect2f	srcImageBox,
		const NvMotDataTypes::Rect2f	tBbox_prev
	)

inline

Recover bounding-box extent clipped by an image border using the previous box size.

Parameters

[in]	tBbox_new	Bounding box that may be clipped by the image border.
[in]	srcImageBox	Source image rectangle whose size defines the clipping border.
[in]	tBbox_prev	Previous bounding box whose width or height is restored on a clipped side.

Returns: Unclipped bounding box.

Recover a left-clipped box.

Recover a right-clipped box.

Recover a top-clipped box.

Recover a bottom-clipped box.

Recover a left-clipped box.

Recover a right-clipped box.

Recover a top-clipped box.

Recover a bottom-clipped box.

Recover a left-clipped box.

Recover a right-clipped box.

Recover a top-clipped box.

Recover a bottom-clipped box.

Definition at line 346 of file sources/libs/nvmultiobjecttracker/include/modules/utility/cv_functions.hpp.

NVIDIA DeepStream SDK API Reference

9.2 Release

Functions

Function Documentation

◆ boundingRect() [1/2]

◆ boundingRect() [2/2]

◆ circshift()

◆ computeIOU()

◆ computeSearchRegion()

◆ computeSizeScore()

◆ computeVisibility()

◆ constructRect2f()

◆ convertBboxToZ()

◆ convertXToBbox()

◆ convexHull() [1/2]

◆ convexHull() [2/2]

◆ displaceBbox()

◆ get2dGaussianPdf()

◆ getBboxCenterPt()

◆ getBoxWithAspectRatio()

◆ getEigenRectString()

◆ getHanningWindow()

◆ isBboxEqual()

◆ modul()

◆ orientation() [1/2]

◆ orientation() [2/2]

◆ unclipBbox()