Functions for detecting corners of various types.

Classes
struct	SearchThreshold
	This class wraps a DetectT class with binary_search_threshold and presents is as a DetectN class. More...
struct	Random
	Detector which randomly scatters corners around an image. More...
struct	DetectN
	A corner detector object which is passed a target number of corners to detect. More...
struct	DetectT
	A corner detector object which is passed a threshold. More...
struct	dog
	Class wrapping the Difference of Gaussians detector. More...
struct	harrisdog
	Class wrapping the Harris-Laplace detector. More...
struct	faster_learn
	FAST-ER detector. More...
struct	ShiTomasiDetect
	Class wrapping the Harris detector. More...
struct	HarrisDetect
	Class wrapping the Shi-Tomasi detector. More...
struct	SUSAN
	Class wrapping the SUSAN detector. More...
Functions
int	binary_search_threshold (const Image< byte > &i, vector< ImageRef > &c, unsigned int N, const DetectT &detector)
auto_ptr< DetectN >	get_detector ()
void	HarrisDetector (const CVD::Image< float > &i, std::vector< std::pair< float, CVD::ImageRef > > &c, unsigned int N, float blur, float sigmas)

---

Function Documentation

int binary_search_threshold	(	const Image< byte > &	i,
		vector< ImageRef > &	c,
		unsigned int	N,
		const DetectT &	detector
	)

This takes a detector which requires a threshold and uses binary search to get as close as possible to the requested number of corners.

Parameters:

	i	The image in which to detect corners.
	c	The detected corners to be returned.
	N	The target number of corners.
	detector	The corner detector.

Definition at line 47 of file detectors.cc.

Referenced by SearchThreshold::operator()().

{
    //Corners for high, low and midpoint thresholds.
    vector<ImageRef> ch, cl, cm;
    
    //The high and low thresholds.
    unsigned int t_high = 256;
    unsigned int t_low = 0;


    detector(i, ch, t_high);
    detector(i, cl, t_low);

    while(t_high > t_low + 1)
    {
    
        cm.clear();
        unsigned int t = (t_high + t_low    ) / 2;
        detector(i, cm, t);

        if(cm.size() == N)
        {
            c = cm;
            return t;
        }
        else if(cm.size() < N) //If we detected too few points, then the t is too high
        {
            t_high = t;
            ch = cm;
        }
        else //We detected too many points to t is too low.
        {
            t_low = t;
            cl = cm;
        }
    }

    //Pick the closest
    //If there is ambiguity, go with the lower threshold (more corners).
    //The only reason for this is that the evaluation code in the FAST-ER
    //system uses this rule.
    if( N - ch.size() >= cl.size() - N)
    {
        c = cl;
        return t_low;
    }
    else
    { 
        c = ch;
        return t_high;
    }
}

auto_ptr<DetectN> get_detector

(

)

Very simple factory function for getting detector objects.

Get a corner detector.

Paramaters (including the detector type) are drawn from the GVars database. The parameter "detector" determines the detector type. Valid options are:

• random Randomly scatter corners around the image.
• dog Difference of Gaussians detector
• harrisdog Harris-Laplace (actually implemented as Harris-DoG) detector
• harris Harris detector with Gaussian blur
• shitomasi Shi-Tomasi detector
• susan Reference implementation of the SUSAN detector
• fast9 libCVD's builtin FAST-9 detector
• fast9old libCVD's builtin FAST-9 detector with the old scoring algorithm, as seen in [Rosten, Drummond 2006].
• fast12 libCVD's builtin FAST-12 detector
• faster2 A FAST-ER detector loaded from a file containing the tree

Definition at line 156 of file detectors.cc.

Referenced by mmain().

{
    
    string d = GV3::get<string>("detector", "fast9", 1);
    
    if(d == "random")
        return auto_ptr<DetectN>(new Random);
    else if(d == "dog")
        return auto_ptr<DetectN>(new dog);
    else if(d == "harrisdog")
        return auto_ptr<DetectN>(new harrisdog);
    else if(d == "shitomasi")
        return auto_ptr<DetectN>(new ShiTomasiDetect);
    else if(d == "harris")
        return auto_ptr<DetectN>(new HarrisDetect);
    #ifdef USESUSAN
        else if(d == "susan")
            return auto_ptr<DetectN>(new SearchThreshold(new SUSAN));
    #endif
    else if(d == "fast9")
        return auto_ptr<DetectN>(new SearchThreshold(new fast_9));
    else if(d == "fast9old")
        return auto_ptr<DetectN>(new SearchThreshold(new fast_9_old));
    else if(d == "fast12")
        return auto_ptr<DetectN>(new SearchThreshold(new fast_12));
    else if(d == "faster2")
        return auto_ptr<DetectN>(new SearchThreshold(new faster_learn(GV3::get<string>("faster2"))));
    else
    {
        cerr << "Unknown detector: " << d << endl;
        exit(1);
    }
}

void HarrisDetector	(	const CVD::Image< float > &	i,
		std::vector< std::pair< float, CVD::ImageRef > > &	c,
		unsigned int	N,
		float	blur,
		float	sigmas
	)

Detect Harris corners.

Parameters:

	i	Image in which to detect corners
	c	Detected corners and strengths are inserted in to this container
	N	Number of corners to detect
	blur	Standard deviation of blur to use
	sigmas	Blur using sigmas standard deviations

Referenced by harrisdog::operator()().

---