4.30 How to detect a bounding box for an object in cv::Mat of OpenCV?
As a simple application of OpenCV APIs, we consider to find a bounding box for an object in cv:Mat of OpenCV. In this example, we have applied the following OpenCV APIs and operations to cv::Mat to find a bounding box.1. Apply cv::EdgePreseringFilter to an originalImage of cv::Mat format.
2. Apply cv::cvtColor to the the filtered Image to get a grayImage.
3. Apply cv::Canny Edge Detector to the grayImage.
4. Apply cv::FindContours to the edgeDetectedImage to find contours.
5. Apply cv::approxPolyDP to the contours to get approximate polygonal coordinates.
6. Get bounding rectangles from the contours polygonal coordinates.
7. Find a single maximum rectangle from the bounding rectanglese set.
For simplicity, we have not used NMS alogrithm in the operation 7 in this program.
The following BoundingBoxDetector is a simple example to implement above operations to detect a single bounding box for an object in cv::Mat image. In practive, this is a very primitive method, but it shows a decent result for an object with a simple background as shown bellow.
/*
* BoundingBoxDetector.cpp
* Copyright (c) 2018 Antillia.com TOSHIYUKI ARAI. ALL RIGHTS RESERVED.
*/
#define _CONSOLE_
#include <sol/ModuleFileName.h>
#include <sol/DropFiles.h>
#include <sol/LabeledTrackBar.h>
#include <sol/FileDialog.h>
#include <sol/StringT.h>
#include <sol/opencv/OpenCVApplicationView.h>
#include <sol/opencv/OpenCVImageView.h>
namespace SOL {
class MainView :public OpenCVApplicationView {
private:
////////////////////////////////////////////////////////////////////////////////////////
//Inner class starts.
class OriginalImageView :public OpenCVImageView {
private:
cv::Mat originalImage;
cv::Mat& getMat()
{
return originalImage;
}
void display()
{
show(originalImage);
}
public:
OriginalImageView(View* parent, const char* name, Args& args)
:OpenCVImageView(parent, name, args)
{
try {
const char* filename = (const char*)args.get(XmNimageFileName);
int imageLoadingFlag = args.get(XmNimageLoadingFlag);
originalImage = readImage(filename, imageLoadingFlag);
} catch (SOL::Exception ex) {
caught(ex);
}
}
~OriginalImageView()
{
}
void loadImage(const char* filename, int imageLoadingFlag=CV_LOAD_IMAGE_COLOR)
{
try {
originalImage = readImage(filename, imageLoadingFlag);
refresh();
} catch (Exception& ex) {
caught(ex);
}
}
};
class DetectedImageView :public OpenCVImageView {
private:
cv::Mat originalImage;
cv::Mat edgePreserved;
cv::Mat grayImage;
cv::Mat cannyEdgedImage;
cv::Mat detectedImage;
cv::Mat& getMat()
{
return detectedImage;
}
void display()
{
show(detectedImage);
}
public:
DetectedImageView(View* parent, const char* name, Args& args)
:OpenCVImageView(parent, name, args)
{
try {
const char* filename = (const char*)args.get(XmNimageFileName);
int imageLoadingFlag = args.get(XmNimageLoadingFlag);
loadImage(filename, imageLoadingFlag);
} catch (SOL::Exception ex) {
caught(ex);
}
}
~DetectedImageView()
{
}
void applyEdgePreservedFilter(double sigmaColor, double sigmaSpace,
cv::Mat& image, cv::Mat& filtered )
{
try {
int flag = RECURS_FILTER ; //NORMCONV_FILTER is very slow;
cv::edgePreservingFilter(
image,
filtered,
flag,
sigmaSpace, //(double)SIGMA_SPACE,
sigmaColor/100.0f); //(double)SIGMA_COLOR/100.0f);
} catch (cv::Exception& ex) {
; //Ignore
}
}
void loadImage(const char* filename, int imageLoadingFlag=CV_LOAD_IMAGE_COLOR)
{
try {
originalImage = readImage( filename, imageLoadingFlag );
detectedImage = originalImage.clone(); //create( originalImage.size(), originalImage.type() );
refresh();
} catch(Exception& ex) {
caught(ex);
}
}
void detectEdge(double sigmaColor, double sigmaSpace, double threshold1, double threshold2)
{
Mat threshold_output;
vector<vector<Point> > contours;
vector<Vec4i> hierarchy;
//1 Apply EdgePreservedFilter to originalImage.
applyEdgePreservedFilter(sigmaColor, sigmaSpace, originalImage, edgePreserved);
//2 Convert it to grayImage.
cv::cvtColor(edgePreserved, grayImage, COLOR_BGR2GRAY );
//3 Apply Canny edge detector to the grayImage.
cv::Canny(grayImage, cannyEdgedImage, threshold1, threshold2);
//4 Find contours from the detectedImage.
cv::findContours(cannyEdgedImage, contours, hierarchy, CV_RETR_TREE, CV_CHAIN_APPROX_SIMPLE, Point(0, 0) );
vector<vector<Point> > contours_poly( contours.size() );
vector<Rect> boundRect( contours.size() );
//5 Get approximate polygonal coordinates from the contours, and
// bounding rectangles from the contours_polygons.
for( int i = 0; i < contours.size(); i++ ) {
cv::approxPolyDP( Mat(contours[i]), contours_poly[i], 3, true );
boundRect[i] = boundingRect( Mat(contours_poly[i]) );
}
detectedImage.release();
detectedImage = originalImage.clone();
//6 Find minimum x and y, and maximum w and h from the boundRect array.
int width = detectedImage.size().width;
int height = detectedImage.size().height;
int x = width, y = height, w = 0, h = 0;
int sx = width, sy = height, ex = 0, ey = 0;
//
for( int i = 0; i< contours.size(); i++ ) {
int xx = boundRect[i].x;
int yy = boundRect[i].y;
int ww = boundRect[i].width;
int hh = boundRect[i].height;
int area = ww * hh;
int MIN_AREA = 10;
int MIN_X = 10;
int MIN_Y = 10;
if (area > MIN_AREA) {
if (xx < sx && xx > MIN_X)
sx = xx;
if (yy < sy && yy > MIN_Y)
sy = yy;
if (xx > ex)
ex = xx;
if (yy > ey)
ey = yy;
}
}
int endx = x + w;
if (endx >= width)
endx = width;
int endy = y + h;
if (endy >= height)
endy = height;
int eex = sx + ex;
int eey = sy + ey;
if (eex > width)
eex = width - 4;
if (eey > height)
eey = height- 4;
//7 Draw a rectangle having a start Point(sx, sy) and end Point(ex, ey) on the contourImage.
cv::rectangle(detectedImage, Point(sx, sy), Point(ex, ey), CV_RGB(255, 0, 0), 2, 8, 0 );
refresh();
}
};
//Inner class ends.
////////////////////////////////////////////////////////////////////////////////////////
StringT<char> imageFile;
StringT<char> savedImageFile;
SmartPtr<OriginalImageView> originalImage;
SmartPtr<DetectedImageView> detectedImage;
static const int SIGMA_COLOR_MAX =200;
int sigmaColor;
SmartPtr<LabeledTrackBar> sigmaColorTrackBar;
static const int SIGMA_SPACE_MAX=100;
int sigmaSpace;
SmartPtr<LabeledTrackBar> sigmaSpaceTrackBar;
static const int THRESHOLD1_MAX = 300;
int threshold1;
SmartPtr<LabeledTrackBar> threshold1TrackBar;
static const int THRESHOLD2_MAX = 300;
int threshold2;
SmartPtr<LabeledTrackBar> threshold2TrackBar;
FileDialog filedlg;
void updateCaption()
{
char caption[MAX_PATH];
sprintf_s(caption, CountOf(caption), "%s - %s",
(const char*)imageFile,
getAppName());
setText(caption);
}
//Horizontal Scroll event by threshold1TrackBar and threshold2TrackBar.
void commonTrackBarCallback(Action& action)
{
sigmaColor = sigmaColorTrackBar -> getPosition();
sigmaSpace = sigmaSpaceTrackBar -> getPosition();
threshold1 = threshold1TrackBar->getPosition();
threshold2 = threshold2TrackBar->getPosition();
detectedImage -> detectEdge((double)sigmaColor, (double)sigmaSpace,
(double)threshold1, (double)threshold2);
}
void resize(int w, int h)
{
if (originalImage && detectedImage &&
sigmaColorTrackBar && sigmaSpaceTrackBar &&
threshold1TrackBar && threshold2TrackBar) {
originalImage -> reshape(2, 2, (w-170)/2-1, h-4);
detectedImage -> reshape((w-160)/2+1, 2, (w-170)/2-1, h-4);
sigmaColorTrackBar -> reshape(w-165 + 10, 10, 155, 70);
sigmaSpaceTrackBar -> reshape(w-165 + 10, 10+70, 155, 70);
threshold1TrackBar -> reshape(w-165 + 10, 10+70*2, 155, 70);
threshold2TrackBar -> reshape(w-165 + 10, 10+70*3, 155, 70);
}
}
void openFile(const char* filename)
{
try {
originalImage -> loadImage(filename);
detectedImage -> loadImage(filename);
const char* fname = strrchr(filename, '\\');
if (fname) {
fname++;
}
imageFile = fname;
updateCaption();
detectedImage -> detectEdge((double)sigmaColor, (double)sigmaSpace,
(double)threshold1, (double)threshold2);
savedImageFile = "";
} catch (Exception& ex) {
caught(ex);
}
}
void dropFiles(Action& action)
{
char fileName[MAX_PATH] = { 0 };
DropFiles drop((HDROP)action.getWParam());
//fileName[0] = ZERO;
int num = drop.queryFile(0, fileName, CountOf(fileName));
if(num > 0) {
if (filedlg.isImageFileName(fileName)) {
openFile(fileName);
bringUp();
} else {
bringUp(); //Activate and raise this view
showErrorDialog("Invalid image filename", fileName, MB_OK);
}
}
}
void confirm(Action& action)
{
int rc = MessageBox(NULL, "Are you sure to close this window?", "Confirmation",
MB_OKCANCEL|MB_ICONEXCLAMATION);
if (rc == IDOK) {
exit(action);
}
}
public:
MainView(OpenCVApplication& applet, const char* name, Args& args)
:OpenCVApplicationView(applet, name, args)
{
imageFile = "..\\images\\CatImage.png";
int imageLoadingFlag = CV_LOAD_IMAGE_COLOR;
try {
Args ar;
ar.set(XmNimageFileName, imageFile);
ar.set(XmNimageLoadingFlag, imageLoadingFlag);
originalImage = new OriginalImageView(this, "cvwindow1", ar);
originalImage -> addCallback(XmNdropCallback, this,
(Callback)&MainView::dropFiles, NULL);
ar.reset();
ar.set(XmNimageFileName, imageFile);
ar.set(XmNimageLoadingFlag, imageLoadingFlag);
detectedImage = new DetectedImageView(this, "cvwindow2", ar);
sigmaColor = 120;
sigmaSpace = 40;
ar.reset();
ar.set(XmNminimum, 0);
ar.set(XmNmaximum, SIGMA_COLOR_MAX);
ar.set(XmNposition, sigmaColor);
ar.set(XmNdisplayOddValue, false);
ar.set(XmNdisplayFloatValue, false);
sigmaColorTrackBar = new LabeledTrackBar(this, "Edge SigmaColor", ar);
sigmaColorTrackBar -> addCallback(XmNtrackBarScrollCallback, this,
(Callback)&MainView::commonTrackBarCallback, NULL);
ar.reset();
ar.set(XmNminimum, 0);
ar.set(XmNmaximum, SIGMA_SPACE_MAX);
ar.set(XmNposition, sigmaSpace);
ar.set(XmNdisplayFloatValue, true);
ar.set(XmNdisplayOddValue, false);
sigmaSpaceTrackBar = new LabeledTrackBar(this, "Edge SigmaSpace", ar);
sigmaSpaceTrackBar -> addCallback(XmNtrackBarScrollCallback, this,
(Callback)&MainView::commonTrackBarCallback, NULL);
threshold1 = 43;
threshold2 = 110;
ar.reset();
ar.set(XmNminimum, 0);
ar.set(XmNmaximum, THRESHOLD1_MAX);
ar.set(XmNposition, threshold1);
ar.set(XmNdisplayOddValue, false);
threshold1TrackBar = new LabeledTrackBar(this, "Canny Threshold1", ar);
threshold1TrackBar -> addCallback(XmNtrackBarScrollCallback, this,
(Callback)&MainView::commonTrackBarCallback, NULL);
ar.reset();
ar.set(XmNminimum, 0);
ar.set(XmNmaximum, THRESHOLD2_MAX);
ar.set(XmNposition, threshold2);
ar.set(XmNdisplayOddValue, false);
threshold2TrackBar = new LabeledTrackBar(this, "Canny Threshold2", ar);
threshold2TrackBar -> addCallback(XmNtrackBarScrollCallback, this,
(Callback)&MainView::commonTrackBarCallback, NULL);
addCallback(XmNmenuCallback, IDM_EXIT, this,
(Callback)&MainView::confirm, NULL);
detectedImage -> detectEdge((double)sigmaColor, (double)sigmaSpace,
(double)threshold1, (double)threshold2);
ar.reset();
ar.set(XmNfilter, FileDialog::getImageFilesFilter());
filedlg.create(this, "OpenFile", ar);
updateCaption();
} catch (Exception& ex) {
caught(ex);
}
}
~MainView()
{
}
void open(Action& action)
{
Args ar;
char dir[MAX_PATH] = { 0 };
if (restoreFileFolder(dir, CountOf(dir))) {
ar.set(XmNdirectory, dir);
filedlg.setValues(ar);
}
try {
if(filedlg.open()) {
const char* filename = filedlg.getFileName();
saveFileFolder(filename);
openFile(filename);
}
} catch (Exception& ex) {
caught(ex);
}
}
//2017/12/01
void save(Action& action)
{
try {
if (!savedImageFile.isEmpty()) {
//Write detected image into the filename.
detectedImage->writeImage(savedImageFile);
} else {
saveAs(action);
}
} catch (Exception& ex) {
caught(ex);
}
}
//2017/12/01
void saveAs(Action& action)
{
Args ar;
char dir[MAX_PATH] = { 0 };
if (restoreFileFolder(dir, CountOf(dir))) {
ar.set(XmNdirectory, dir);
filedlg.setValues(ar);
}
try {
if(filedlg.save()) {
const char* filename = filedlg.getFileName();
saveFileFolder(filename);
//Write detected image into the filename.
detectedImage->writeImage(filename);
savedImageFile = filename;
}
} catch (Exception& ex) {
caught(ex);
}
}
};
}
//
void main(int argc, char** argv)
{
try {
ModuleFileName module(argv[0]);
const char* name = module.getAppName();
OpenCVApplication applet(name, argc, argv);
Args args;
args.set(XmNwidth, 860);
args.set(XmNheight, 360);
MainView view(applet, name, args);
view.realize();
applet.run();
} catch (SOL::Exception& ex) {
caught(ex);
}
}
Last modified: 15 Apr. 2018
Copyright (c) 2018 Antillia.com ALL RIGHTS RESERVED.