HOG represents a class for computing the HOG descriptor described in Dalal, N. and Triggs, B., "Histograms of oriented gradients for human detection", CVPR 2005. More...

#include <hog.h>

Public Member Functions
void	compute (float I, int h, int w, int n_channels, int bin_size, int n_orients, bool soft_bin, float descriptor)
	Compute HOG descriptor.
void	compute (float I, float descriptor) const
	Compute HOG descriptor with default parameters.
void	gradHist (float M, float O, int h, int w, int bin_size, int n_orients, bool soft_bin, float *H) const
	Compute n_orients gradient histograms per bin_size x bin_size block of pixels.
void	gradMag (float I, int h, int w, int d, float M, float *O) const
	Compute gradient magnitude and orientation at each location (uses sse).
	HOG ()
	Constructor.
void	normalization (float H, int h, int w, int bin_size, int n_orients, float clip, float G) const
	Normalize histogram of gradients.
virtual	~HOG ()
	Destructor.
Protected Attributes
int	bin_size_
	spatial bin size (default = 8)
float	clip_
	value at which to clip histogram bins (default = 0.2)
int	h_
	image height (default = 128)
int	n_channels_
	image number of channels (default = 3)
int	n_orients_
	number of orientation bins (default = 9)
bool	soft_bin_
	if true, each pixel can contribute to multiple spatial bins (using bilinear interpolation) (default = true)
int	w_
	image width (default = 64)
Private Member Functions
float *	acosTable () const
	Build lookup table a[] s.t. a[dx/2.02*n]~=acos(dx).
void	alFree (void *aligned) const
	Platform independent aligned memory de-allocation (see also alMalloc).
void *	alMalloc (size_t size, int alignment) const
	Platform independent aligned memory allocation (see also alFree).
void	grad1 (float I, float Gx, float *Gy, int h, int w, int x) const
	Compute x and y gradients for just one column (uses sse).
void	gradQuantize (float O, float M, int O0, int O1, float M0, float M1, int n_orients, int nb, int n, float norm) const
	Helper for gradHist, quantize O and M into O0, O1 and M0, M1 (uses sse).

Detailed Description

HOG represents a class for computing the HOG descriptor described in Dalal, N. and Triggs, B., "Histograms of oriented gradients for human detection", CVPR 2005.

Author:: Matteo Munaro, Stefano Ghidoni, Stefano Michieletto

Definition at line 55 of file hog.h.

Constructor & Destructor Documentation

pcl::people::HOG::HOG ( )

Constructor.

Definition at line 55 of file hog.cpp.

pcl::people::HOG::~HOG ( ) [virtual]

Destructor.

Definition at line 68 of file hog.cpp.

Member Function Documentation

float * pcl::people::HOG::acosTable ( ) const [private]

Build lookup table a[] s.t. a[dx/2.02*n]~=acos(dx).

Definition at line 419 of file hog.cpp.

void pcl::people::HOG::alFree ( void * aligned ) const [inline, private]

Platform independent aligned memory de-allocation (see also alMalloc).

Definition at line 488 of file hog.cpp.

void * pcl::people::HOG::alMalloc	(	size_t	size,
		int	alignment
	)		const `[inline, private]`

Platform independent aligned memory allocation (see also alFree).

Definition at line 478 of file hog.cpp.

void pcl::people::HOG::compute	(	float *	I,
		int	h,
		int	w,
		int	n_channels,
		int	bin_size,
		int	n_orients,
		bool	soft_bin,
		float *	descriptor
	)

Compute HOG descriptor.

Parameters:

[in]	I	Image as array of float between 0 and 1.
[in]	h	Image height.
[in]	w	Image width.
[in]	n_channels	Image number of channels.
[in]	bin_size	Spatial bin size.
[in]	n_orients	Number of orientation bins.
[in]	soft_bin	If true, each pixel can contribute to multiple spatial bins (using bilinear interpolation).
[out]	descriptor	HOG descriptor.

Definition at line 309 of file hog.cpp.

void pcl::people::HOG::compute	(	float *	I,
		float *	descriptor
	)		const

Compute HOG descriptor with default parameters.

Parameters:

[in]	I	Image as array of float between 0 and 1.
[out]	descriptor	HOG descriptor.

Definition at line 322 of file hog.cpp.

void pcl::people::HOG::grad1	(	float *	I,
		float *	Gx,
		float *	Gy,
		int	h,
		int	w,
		int	x
	)		const `[private]`

Compute x and y gradients for just one column (uses sse).

Definition at line 357 of file hog.cpp.

void pcl::people::HOG::gradHist	(	float *	M,
		float *	O,
		int	h,
		int	w,
		int	bin_size,
		int	n_orients,
		bool	soft_bin,
		float *	H
	)		const

Compute n_orients gradient histograms per bin_size x bin_size block of pixels.

Parameters:

[in]	M	Gradient magnitude for each image point.
[in]	O	Gradient orientation for each image point.
[in]	h	Image height.
[in]	w	Image width.
[in]	bin_size	Spatial bin size.
[in]	n_orients	Number of orientation bins.
[in]	soft_bin	If true, each pixel can contribute to multiple spatial bins (using bilinear interpolation).
[out]	H	Gradient histograms.

Definition at line 171 of file hog.cpp.

void pcl::people::HOG::gradMag	(	float *	I,
		int	h,
		int	w,
		int	d,
		float *	M,
		float *	O
	)		const

Compute gradient magnitude and orientation at each location (uses sse).

Parameters:

[in]	I	Image as array of float.
[in]	h	Image height.
[in]	w	Image width.
[in]	d	Image number of channels.
[out]	M	Gradient magnitude for each image point.
[out]	O	Gradient orientation for each image point.

Definition at line 71 of file hog.cpp.

void pcl::people::HOG::gradQuantize	(	float *	O,
		float *	M,
		int *	O0,
		int *	O1,
		float *	M0,
		float *	M1,
		int	n_orients,
		int	nb,
		int	n,
		float	norm
	)		const `[private]`

Helper for gradHist, quantize O and M into O0, O1 and M0, M1 (uses sse).

Definition at line 434 of file hog.cpp.

void pcl::people::HOG::normalization	(	float *	H,
		int	h,
		int	w,
		int	bin_size,
		int	n_orients,
		float	clip,
		float *	G
	)		const

Normalize histogram of gradients.

Parameters:

[in]	H	Gradient histograms.
[in]	h	Image height.
[in]	w	Image width.
[in]	bin_size	Spatial bin size.
[in]	n_orients	Number of orientation bins.
[in]	clip	Value at which to clip histogram bins.
[out]	G	Normalized gradient histograms.