A linear classifier for HOG features. More...

#include <ODHOGDetector.h>

Inheritance diagram for od::g2d::ODHOGDetector:

Collaboration diagram for od::g2d::ODHOGDetector:

Public Member Functions
	OD_DEFINE_ENUM_WITH_STRING_CONVERSIONS (SVMType,(OD_CUSTOM)(OD_DEFAULT_PEOPLE)(OD_DAIMLER_PEOPLE)(OD_FILE)) ODHOGDetector(std

void	init ()

void	load (std::string filename)

void	setSVMFromFile (std::string fileName)

void	setSVMDetector (std::vector< float > svmdetector)

ODDetections2D *	detectOmni (ODSceneImage *scene)
	Function for performing detection on an entire scene. More...

ODDetections *	detect (ODSceneImage *scene)
	Function for performing detection on a segmented scene. More...

int	detect (ODScene scene, std::vector< ODDetection > &detections)

void	setTrainedDataLocation (std::string trained_data_location_)
	The base directory for trained data. More...

SVMType const &	getSvmtype () const

void	setSvmtype (SVMType const &svmtype_)

cv::Size const &	getWinSize () const

void	setWinSize (cv::Size const &winSize)

cv::Size const &	getBlockSize () const

void	setBlockSize (cv::Size const &blockSize)

cv::Size const &	getBlockStride () const

void	setBlockStride (cv::Size const &blockStride)

cv::Size const &	getCellSize () const

void	setCellSize (cv::Size const &cellSize)

float	getHitThreshold () const

void	setHitThreshold (float hitThreshold)

void	printParameters ()

Public Member Functions inherited from od::ODDetector2D
	ODDetector2D (std::string const &trained_data_location_)

ODDetections *	detect (ODScene *scene)

Public Member Functions inherited from od::ODDetector
	ODDetector (std::string const &training_data_location_)

virtual ODDetections *	detectOmni (ODScene *scene)

Public Member Functions inherited from od::ODDetectorCommon
	ODDetectorCommon (std::string const &trained_data_location_="")

std::string	getTrainingInputLocation () const
	Gets/Sets the directory containing the data for training. More...

void	setTrainingInputLocation (std::string training_input_location_)
	Gets/Sets the directory containing the data for training. More...

std::string	getTrainedDataLocation () const
	Gets/Sets the base directory for trained data. More...

std::string	getSpecificTrainingDataLocation ()
	Gets the specific directory for a Trainer or a Detector inside trained_data_location_. More...

std::string	getSpecificTrainingData ()

std::string const &	getTrainedDataID () const

void	setTrainedDataID (std::string const &trainedDataID)

Protected Attributes
cv::Size	winSize

cv::Size	blockSize

cv::Size	blockStride

cv::Size	cellSize

float	hitThreshold

cv::HOGDescriptor	hog_

SVMType	svmtype_

Protected Attributes inherited from od::ODDetectorCommon
std::string	training_input_location_

std::string	trained_data_location_

std::string	TRAINED_DATA_ID_

std::string	TRAINED_LOCATION_DENTIFIER_

Additional Inherited Members
Public Attributes inherited from od::ODDetector
bool	metainfo_

Detailed Description

A linear classifier for HOG features.

This class takes an image as an input, finds its HOG features using the parameters containing in the trained data, runs the linear classifier (present in the trained data) on the computed HOG features and finally produces the classification output. Covers both multiscale detection (detectOmni()) and detection on a fixed scene (detect() - the scene is resized to the HOG window size).

Use ODHOGTrainer for the training of new objects. By default this class provides two people detector: OD_DEFAULT_PEOPLE, the people detector from OpenCV and OD_DAIMLER_PEOPLE, the second detector available from OpenCV. One can set any linear classifier, the linear weight vector through setSVMDetector() function but it is highly not recommended as the HOG parameters and the weight vector can be out of sync. One must set appropriate HOG parameters after using setSVMDetector() function with which the weight vector was trained.

Author: Kripasindhu Sarkar

Examples:: apps/global2D/od_multihog_app.cpp, objectdetector/od_hog_train.cpp, and objectdetector/od_image_hog_files.cpp.

Definition at line 60 of file ODHOGDetector.h.

Member Function Documentation

ODDetections* od::g2d::ODHOGDetector::detect ( ODSceneImage * scene )

virtual

Function for performing detection on a segmented scene.

The purpose of this function is to perform detection on a segmented scene or an 'object candidate'. i.e. the entire scene is considered as an 'object' or an detection. It is possible for a scene to trigger multiple detections.

Parameters

[in] scene An instance of 2D scene

Returns: [out] detections A number of detections as an ODDetections instance.

Implements od::ODDetector2D.

int od::g2d::ODHOGDetector::detect	(	ODScene *	scene,
		std::vector< ODDetection * > &	detections
	)

inline

Definition at line 95 of file ODHOGDetector.h.

ODDetections2D* od::g2d::ODHOGDetector::detectOmni ( ODSceneImage * scene )

virtual

Function for performing detection on an entire scene.

The purpose of this function is to detect an object in an entire scene. Thus, other than the type of detection we also have information about the location of the detection w.r.t. the scene.

Parameters

[in] scene An instance of 2D scene

Returns: [out] detections A number of detections as an ODDetections2D instance containing information about the detection and its 2D location.

Implements od::ODDetector2D.

Examples:: objectdetector/od_hog_train.cpp, and objectdetector/od_image_hog_files.cpp.

cv::Size const& od::g2d::ODHOGDetector::getBlockSize ( ) const

inline

Definition at line 125 of file ODHOGDetector.h.

References blockSize.

cv::Size const& od::g2d::ODHOGDetector::getBlockStride ( ) const

inline

Definition at line 135 of file ODHOGDetector.h.

References blockStride.

cv::Size const& od::g2d::ODHOGDetector::getCellSize ( ) const

inline

Definition at line 145 of file ODHOGDetector.h.

References cellSize.

float od::g2d::ODHOGDetector::getHitThreshold ( ) const

inline

Definition at line 155 of file ODHOGDetector.h.

References hitThreshold.

SVMType const& od::g2d::ODHOGDetector::getSvmtype ( ) const

inline

Definition at line 105 of file ODHOGDetector.h.

References svmtype_.

cv::Size const& od::g2d::ODHOGDetector::getWinSize ( ) const

inline

Definition at line 115 of file ODHOGDetector.h.

References winSize.

void od::g2d::ODHOGDetector::init ( )

virtual

Implements od::ODDetectorCommon.

Examples:: objectdetector/od_hog_train.cpp, and objectdetector/od_image_hog_files.cpp.

void od::g2d::ODHOGDetector::load ( std::string filename )

od::g2d::ODHOGDetector::OD_DEFINE_ENUM_WITH_STRING_CONVERSIONS	(	SVMType	,
		(OD_CUSTOM)(OD_DEFAULT_PEOPLE)(OD_DAIMLER_PEOPLE)(OD_FILE)
	)

inline

Definition at line 64 of file ODHOGDetector.h.

References od::ODDetector::metainfo_, svmtype_, od::ODDetectorCommon::TRAINED_DATA_ID_, and od::ODDetectorCommon::TRAINED_LOCATION_DENTIFIER_.

void od::g2d::ODHOGDetector::printParameters ( )

void od::g2d::ODHOGDetector::setBlockSize ( cv::Size const & blockSize )

inline

Definition at line 130 of file ODHOGDetector.h.

References blockSize.

void od::g2d::ODHOGDetector::setBlockStride ( cv::Size const & blockStride )

inline

Definition at line 140 of file ODHOGDetector.h.

References blockStride.

void od::g2d::ODHOGDetector::setCellSize ( cv::Size const & cellSize )

inline

Definition at line 150 of file ODHOGDetector.h.

References cellSize.

void od::g2d::ODHOGDetector::setHitThreshold ( float hitThreshold )

inline

Definition at line 160 of file ODHOGDetector.h.

References hitThreshold.

void od::g2d::ODHOGDetector::setSVMDetector ( std::vector< float > svmdetector )

inline

Definition at line 87 of file ODHOGDetector.h.

References hog_.

void od::g2d::ODHOGDetector::setSVMFromFile ( std::string fileName )

void od::g2d::ODHOGDetector::setSvmtype ( SVMType const & svmtype_ )

inline

Examples:: apps/global2D/od_multihog_app.cpp.

Definition at line 110 of file ODHOGDetector.h.

References svmtype_.

void od::g2d::ODHOGDetector::setTrainedDataLocation ( std::string trained_data_location_ )

inlinevirtual

The base directory for trained data.

This should be same for all Trainers and Detectors and can be considered as the 'database' of trained data. Trainers uses one of its subdirectories based on its type to store algo specific trained data. The corresponding Detector would use the same directory to fetch the trained data for online detection.

Reimplemented from od::ODDetectorCommon.