Detection< A > Struct Template Reference

The Detection structure is a high-level C++ wrapper for object detection with HAAR and LBP cascade classifiers. More...

#include <SimdDetection.hpp>

Data Structures
struct	Object
	The Object structure describes one grouped detection result. More...

Public Types
typedef A< uint8_t >	Allocator
typedef Simd::View< A >	View
typedef Simd::Point< ptrdiff_t >	Size
typedef std::vector< Size >	Sizes
typedef Simd::Rectangle< ptrdiff_t >	Rect
typedef std::vector< Rect >	Rects
typedef int	Tag
typedef std::vector< Object >	Objects

Public Member Functions
	Detection ()
	~Detection ()
bool	LoadStringXml (const std::string &xml, Tag tag=UNDEFINED_OBJECT_TAG)
bool	Load (const std::string &path, Tag tag=UNDEFINED_OBJECT_TAG)
bool	Init (const Size &imageSize, double scaleFactor=1.1, const Size &sizeMin=Size(0, 0), const Size &sizeMax=Size(INT_MAX, INT_MAX), const View &roi=View(), ptrdiff_t threadNumber=-1)
bool	Detect (const View &src, Objects &objects, int groupSizeMin=3, double sizeDifferenceMax=0.2, bool motionMask=false, const Rects &motionRegions=Rects())

Static Public Attributes
static const Tag	UNDEFINED_OBJECT_TAG = -1

Detailed Description

template<template< class > class A>
struct Simd::Detection< A >

The Detection structure is a high-level C++ wrapper for object detection with HAAR and LBP cascade classifiers.

The structure loads one or more OpenCV-format cascades, prepares a scale pyramid for a fixed input image size and then detects objects in images of this size. The input image can be gray or color; color images are converted to gray internally. If several cascades are loaded, every cascade is applied independently and detections are grouped separately for every user tag.

Using example (face detection in the image):

#include "Simd/SimdDetection.hpp"
#include "Simd/SimdDrawing.hpp"
 
int main()
{
    typedef Simd::Detection<Simd::Allocator> Detection;
 
    Detection::View image;
    image.Load("../../data/image/face/lena.pgm");
 
    Detection detection;
 
    detection.Load("../../data/cascade/haar_face_0.xml");
 
    detection.Init(image.Size());
 
    Detection::Objects objects;
    detection.Detect(image, objects);
 
    for (size_t i = 0; i < objects.size(); ++i)
        Simd::DrawRectangle(image, objects[i].rect, uint8_t(255));
 
    image.Save("result.pgm");
 
    return 0;
}

Using example (face detection in the video captured by OpenCV):

#include <iostream>
#include <string>
 
#include "opencv2/opencv.hpp"
#include "opencv2/core/utils/logger.hpp"
#ifndef SIMD_OPENCV_ENABLE
#define SIMD_OPENCV_ENABLE
#endif
#include "Simd/SimdDetection.hpp"
#include "Simd/SimdDrawing.hpp"
 
int main(int argc, char * argv[])
{
    if (argc < 2)
    {
        std::cout << "You have to set video source! It can be 0 for camera or video file name." << std::endl;
        return 1;
    }
    std::string source = argv[1], output = argc > 2 ? argv[2] : "";
 
    cv::VideoCapture capture;
    cv::utils::logging::setLogLevel(cv::utils::logging::LOG_LEVEL_ERROR);
    if (source == "0")
        capture.open(0);
    else
        capture.open(source);
    if (!capture.isOpened())
    {
        std::cout << "Can't capture '" << source << "' !" << std::endl;
        return 1;
    }
    int W = (int)capture.get(cv::CAP_PROP_FRAME_WIDTH);
    int H = (int)capture.get(cv::CAP_PROP_FRAME_HEIGHT);
 
    cv::VideoWriter writer;
    if (output.size())
    {
        writer.open(output, cv::VideoWriter::fourcc('F', 'M', 'P', '4'), capture.get(cv::CAP_PROP_FPS), cv::Size(W, H));
        if (!writer.isOpened())
        {
            std::cout << "Can't open output file '" << output << "' !" << std::endl;
            return 1;
        }
    }
 
    typedef Simd::Detection<Simd::Allocator> Detection;
    Detection detection;
    detection.Load("../../data/cascade/haar_face_0.xml");
    detection.Init(Detection::Size(W, H), 1.2, Detection::Size(W, H) / 20);
 
    const char * WINDOW_NAME = "FaceDetection";
    cv::namedWindow(WINDOW_NAME, 1);
    for (;;)
    {
        cv::Mat frame;
        if (!capture.read(frame))
            break;
        Detection::View image = frame;
 
        Detection::Objects objects;
        detection.Detect(image, objects);
 
        for (size_t i = 0; i < objects.size(); ++i)
            Simd::DrawRectangle(image, objects[i].rect, Simd::Pixel::Bgr24(0, 255, 255));
 
        cv::imshow(WINDOW_NAME, frame);
        if (writer.isOpened())
            writer.write(frame);
        if (cv::waitKey(1) == 27)// "press 'Esc' to break video";
            break;
    }
    return 0;
}

Note

This is wrapper around low-level Object Detection API.

Member Typedef Documentation

Allocator

typedef A<uint8_t> Allocator

Allocator used by temporary images and buffers.

View

typedef Simd::View<A> View

Image view type used for input images, ROI masks and internal pyramid levels.

Size

typedef Simd::Point<ptrdiff_t> Size

Two-dimensional size or point type.

Sizes

typedef std::vector<Size> Sizes

Vector of sizes.

Rect

typedef Simd::Rectangle<ptrdiff_t> Rect

Rectangle type used for object bounds and search regions.

Rects

typedef std::vector<Rect> Rects

Vector of rectangles.

Tag

typedef int Tag

User tag type used to identify detections from different cascades.

Objects

typedef std::vector<Object> Objects

Vector of detected objects.

Constructor & Destructor Documentation

Detection()

Detection

(

)

Creates a new empty Detection structure. Load at least one cascade and call Init before Detect.

~Detection()

~Detection

(

)

Releases all loaded cascades.

Member Function Documentation

LoadStringXml()

bool LoadStringXml	(	const std::string &	xml,
		Tag	tag = UNDEFINED_OBJECT_TAG
	)

Loads a classifier cascade from XML text. Supports OpenCV HAAR and LBP cascade types. You can call this function more than once if you want to use several object detectors at the same time.

Note

Tree based cascades and old cascade formats are not supported!

Parameters

[in]	xml	- a string containing XML with cascade.
[in]	tag	- a user defined tag. This tag will be inserted into output Object structures.

Returns

a result of this operation.

Load()

bool Load	(	const std::string &	path,
		Tag	tag = UNDEFINED_OBJECT_TAG
	)

Loads a classifier cascade from an XML file. Supports OpenCV HAAR and LBP cascade types. You can call this function more than once if you want to use several object detectors at the same time.

Note

Tree based cascades and old cascade formats are not supported!

Parameters

[in]	path	- a path to XML cascade file.
[in]	tag	- a user defined tag. This tag will be inserted into output Object structures.

Returns

a result of this operation.

Init()

bool Init	(	const Size &	imageSize,
		double	scaleFactor = 1.1,
		const Size &	sizeMin = Size(0, 0),
		const Size &	sizeMax = Size(INT_MAX, INT_MAX),
		const View &	roi = View(),
		ptrdiff_t	threadNumber = -1
	)

Prepares Detection structure to work with images of the given size.

The function builds internal pyramid levels for all loaded cascades and object sizes that fit into [sizeMin, sizeMax]. It must be called after loading cascades and before Detect. After successful initialization every input image passed to Detect must have the same size as imageSize.

Parameters

[in]	imageSize	- a size of input image.
[in]	scaleFactor	- a multiplier between neighboring pyramid levels. Values close to 1 improve scale precision but create more levels and reduce performance.
[in]	sizeMin	- a minimal size of detected objects in input image coordinates. It strongly affects performance.
[in]	sizeMax	- a maximal size of detected objects in input image coordinates.
[in]	roi	- an optional 8-bit mask which defines a Region Of Interest. Non-zero mask pixels allow detection; zero pixels reject it. The mask is applied to detected object centers.
[in]	threadNumber	- a number of worker threads. Use value -1 to choose this number automatically.

Returns

a result of this operation.

Detect()

bool Detect	(	const View &	src,
		Objects &	objects,
		int	groupSizeMin = 3,
		double	sizeDifferenceMax = 0.2,
		bool	motionMask = false,
		const Rects &	motionRegions = Rects()
	)

Detects objects in the given image.

The function runs all initialized pyramid levels, collects elementary detections and then groups similar rectangles separately for every cascade tag. The output vector is cleared before grouped objects are added.

Parameters

[in]	src	- an input image. Its size must be equal to imageSize passed to Init.
[out]	objects	- detected and grouped objects.
[in]	groupSizeMin	- a minimal number of elementary detections required to keep a grouped object. If this value is zero, no objects are returned.
[in]	sizeDifferenceMax	- a relative rectangle difference used to group elementary detections.
[in]	motionMask	- a flag to restrict detection by motionRegions in addition to the ROI mask.
[in]	motionRegions	- rectangles in input image coordinates that dynamically restrict detection. The regions are applied to detected object centers.

Returns

a result of this operation.

Field Documentation

UNDEFINED_OBJECT_TAG

const Tag UNDEFINED_OBJECT_TAG = -1

static

Default tag assigned to objects when no user tag is specified.