The main class. More...

#include <MultiObjectTLD.h>

Public Member Functions
void	addObject (ObjectBox b)
	Marks a new object in the previously passed frame.
void	addObjects (std::vector< ObjectBox > obs)
	Adds multiple objects in the previously passed frame.
void	enableLearning (bool enable=true)
	En/Disables learning (i.e. updating the classifiers) at runtime.
void	getDebugImage (unsigned char *src, Matrix &rMat, Matrix &gMat, Matrix &bMat, int mode=255) const
	Writes a colored debug image into the given rgb matrices. Details see writeDebugImage().
ObjectBox	getObjectBox () const
	Returns the location of the first object.
std::vector< ObjectBox >	getObjectBoxes () const
	Returns the current object positions.
int	getStatus (const int objId=0) const
	Returns current status of object `objId`.
bool	getValid () const
	Returns if the first object is valid (equivalent to `getStatus(0)` returning `STATUS_OK`)
	MultiObjectTLD (const int width, const int height, const MOTLDSettings &settings=MOTLDSettings())
	Default constructor.
void	processFrame (unsigned char *img)
	Processes the current frame (tracking - detecting - learning)
void	saveClassifier (const char *filename) const
	Saves the classifier to a (binary) file.
void	writeDebugImage (unsigned char src, char filename, int mode=255) const
	Saves an output image to file in PPM format.
Static Public Member Functions
static MultiObjectTLD	loadClassifier (const char *filename)
	Returns an instance of MultiObjectTLD while loading the classifier from file.
Private Member Functions
void	clusterDetections (float threshold)
	MultiObjectTLD (int width, int height, int colorMode, int patchSize, int bbMin, bool useColor, bool fastRotation, NNClassifier nnc, FernFilter ff, int nObjects, float aspectRatio, bool learningEnabled)
Private Attributes
float	ivAspectRatio
int	ivBBmin
int	ivColorMode
Matrix	ivCurImage
unsigned char *	ivCurImagePtr
std::vector< ObjectBox >	ivCurrentBoxes
std::vector< NNPatch >	ivCurrentPatches
std::vector< bool >	ivDefined
bool	ivEnableFastRotation
FernFilter	ivFernFilter
int	ivHeight
std::vector< FernDetection >	ivLastDetectionClusters
std::vector< FernDetection >	ivLastDetections
bool	ivLearningEnabled
LKTracker	ivLKTracker
int	ivNLastDetections
NNClassifier	ivNNClassifier
int	ivNObjects
int	ivPatchSize
bool	ivUseColor
std::vector< bool >	ivValid
int	ivWidth

Detailed Description

The main class.

Definition at line 92 of file MultiObjectTLD.h.

Constructor & Destructor Documentation

motld::MultiObjectTLD::MultiObjectTLD	(	const int	width,
		const int	height,
		const MOTLDSettings &	settings = `MOTLDSettings()`
	)		`[inline]`

Default constructor.

Parameters:

width	width...
height	...and height of the images in the sequence
settings	configuration structure (see class MOTLDSettings for more details)

Definition at line 100 of file MultiObjectTLD.h.

motld::MultiObjectTLD::MultiObjectTLD	(	int	width,
		int	height,
		int	colorMode,
		int	patchSize,
		int	bbMin,
		bool	useColor,
		bool	fastRotation,
		NNClassifier	nnc,
		FernFilter	ff,
		int	nObjects,
		float	aspectRatio,
		bool	learningEnabled
	)		`[private]`

Definition at line 756 of file MultiObjectTLD.h.

Member Function Documentation

void motld::MultiObjectTLD::addObject ( ObjectBox b )

Marks a new object in the previously passed frame.

Note:: To add multiple objects in a single frame please prefer addObjects().

Definition at line 201 of file MultiObjectTLD.h.

void motld::MultiObjectTLD::addObjects ( std::vector< ObjectBox > obs )

Adds multiple objects in the previously passed frame.

Note:: For efficiency reasons all object boxes need to have the same aspect ratio. All further boxes are automatically reshaped to the aspect ratio of the first object box while preserving the area.

Definition at line 208 of file MultiObjectTLD.h.

void motld::MultiObjectTLD::clusterDetections ( float threshold ) [private]

Definition at line 538 of file MultiObjectTLD.h.

void motld::MultiObjectTLD::enableLearning ( bool enable = true ) [inline]

En/Disables learning (i.e. updating the classifiers) at runtime.

Definition at line 127 of file MultiObjectTLD.h.

void motld::MultiObjectTLD::getDebugImage	(	unsigned char *	src,
		Matrix &	rMat,
		Matrix &	gMat,
		Matrix &	bMat,
		int	mode = `255`
	)		const

Writes a colored debug image into the given rgb matrices. Details see writeDebugImage().

Definition at line 660 of file MultiObjectTLD.h.

ObjectBox motld::MultiObjectTLD::getObjectBox ( ) const [inline]

Returns the location of the first object.

Definition at line 139 of file MultiObjectTLD.h.

std::vector<ObjectBox> motld::MultiObjectTLD::getObjectBoxes ( ) const [inline]

Returns the current object positions.

Definition at line 141 of file MultiObjectTLD.h.

int motld::MultiObjectTLD::getStatus ( const int objId = 0 ) const

Returns current status of object objId.

Returns:

one of the following states:

STATUS_OK: the object is considered to be found correctly in the current frame
STATUS_UNSURE: the object may be lost, but is still tracked
STATUS_LOST: the object is definitely lost (the corresponding box returned by getObjectBoxes() is invalid)

Definition at line 264 of file MultiObjectTLD.h.

bool motld::MultiObjectTLD::getValid ( ) const [inline]

Returns if the first object is valid (equivalent to getStatus(0) returning STATUS_OK)

Definition at line 137 of file MultiObjectTLD.h.

MultiObjectTLD motld::MultiObjectTLD::loadClassifier ( const char * filename ) [static]

Returns an instance of MultiObjectTLD while loading the classifier from file.

Definition at line 805 of file MultiObjectTLD.h.

void motld::MultiObjectTLD::processFrame ( unsigned char * img )

Processes the current frame (tracking - detecting - learning)

Parameters:

img The image passed as an unsigned char array. The pixels are assumed to be given row by row from top left to bottom right with the image width and height passed to the constructor (see MultiObjectTLD()). In case of COLOR_MODE_RGB the array should have the form [r_0, r_1, ..., r_n, g_0, g_1, ..., g_n, b_0, b_1, ..., b_n].

Definition at line 275 of file MultiObjectTLD.h.

void motld::MultiObjectTLD::saveClassifier ( const char * filename ) const

Saves the classifier to a (binary) file.

Definition at line 777 of file MultiObjectTLD.h.

void motld::MultiObjectTLD::writeDebugImage	(	unsigned char *	src,
		char *	filename,
		int	mode = `255`
	)		const

Saves an output image to file in PPM format.

Parameters:

src	the same as passed to processFrame()
filename	the filename, suggested ending: ".ppm"
mode	controls which components are drawn, bitwise or of DEBUG_DRAW_DETECTIONS, DEBUG_DRAW_PATCHES, DEBUG_DRAW_CROSSES.

The patches on the left side represent negative examples, positive examples are on the right side, starting a new column for each object. The red bars next to the patches sketch the corresponding color histograms. Green boxes represent detections shortlisted by the filtering process (see FernFilter). Blue boxes represent cluster means (if enabled). Blue crosses represent tracking points that are considered as inliers (see LKTracker). A red box stands for STATUS_OK, a yellow box means STATUS_UNSURE (cf. getStatus()).

Definition at line 651 of file MultiObjectTLD.h.