![Getting Images from the Webcam
#include <stdio.h>
#include "opencv/cv.h"
#include "opencv/highgui.h"
#include "opencv/cxcore.h"
int main(int argc, char* argv[])
{
CvCapture *capture = 0; //The camera
IplImage* frame = 0; //The images you bring out of the camera
capture = cvCaptureFromCAM( 0 ); //Open the camera
if (!capture ){
printf("Could not connect to cameran" ); return 1;
}
cvNamedWindow( "demowin1", CV_WINDOW_AUTOSIZE ); //Create output window
while(key != 27 /*escape key to quit*/)
{
//Query for the next frame
frame = cvQueryFrame( capture );
if( !frame ) break;
//show the raw image in the first window
cvShowImage( "demowin1", frame );
//Get the last key that's been pressed for input
key = cvWaitKey( 1 );
}
//Shutdown
cvReleaseCapture( &capture ); // WARNING: DO NOT release the "frame" variable
cvDestroyWindow( "demowin1" );
return 0;
}](https://image.slidesharecdn.com/opencvtutorial-120522223338-phpapp01/85/Open-cv-tutorial-9-320.jpg)
![Saving a video file
int main(int argc, char* argv[])
{
CvCapture *capture = 0; //The camera
IplImage* frame = 0; //The images you bring out of the camera
capture = cvCaptureFromCAM( 0 ); //Open the camera
if (!capture ){
printf("Could not connect to cameran" ); return 1;
}
CvVideoWriter *writer;
int isColor = 1;
int fps = 25; // or 30
writer = cvCreateVideoWriter("output.avi", CV_FOURCC_DEFAULT, fps, cvSize(640, 480));
while(key != 27 /*escape key to quit*/)
{
//Query for the next frame
frame = cvQueryFrame( capture );
if( !frame ) break;
cvWriteFrame(writer, frame); //Just like showing an image
//Get the last key that's been pressed for input
key = cvWaitKey( 1 );
}
cvReleaseVideoWriter(&writer); //Release the video file
cvReleaseCapture( &capture ); // WARNING: DO NOT release the "frame" variable
return 0;
}](https://image.slidesharecdn.com/opencvtutorial-120522223338-phpapp01/85/Open-cv-tutorial-10-320.jpg)
![Accessing a pixel
{
int i = frame->height/2;
int j = frame->width/2;
//slow way
CvScalar bgr = cvGet2D(frame, i, j);
printf("B:%d, G:%d, R:%dn", (int)bgr.val[0], (int)bgr.val[1], (int)bgr.val[2]);
//fast way (access the array directly)
printf("B:%d, G:%d, R:%dn",
((uchar *)(frame->imageData + i*frame->widthStep))[j*frame->nChannels + 0], // B
((uchar *)(frame->imageData + i*frame->widthStep))[j*frame->nChannels + 1], // G
((uchar *)(frame->imageData + i*frame->widthStep))[j*frame->nChannels + 2]);// R
//set the value
cvSet2D(frame, i, j, CV_RGB(0,255,0));
cvShowImage( "demowin2", frame );
}](https://image.slidesharecdn.com/opencvtutorial-120522223338-phpapp01/85/Open-cv-tutorial-11-320.jpg)
![2-1. Fill the fragmented pixles
1. Background
Video Model
Source Construction
2. Foreground Object 3. Upper body 4. Distance 5. Silhouette
Extraction Identification Transform extraction
Closing by Reconstruction [5]
24](https://image.slidesharecdn.com/opencvtutorial-120522223338-phpapp01/85/Open-cv-tutorial-24-320.jpg)
![2-2. Shadow detection [4]
1. Background
Video Model
Source Construction
2. Foreground Object 3. Upper body 4. Distance 5. Silhouette
Extraction Identification Transform extraction
R
G
Shadow pixel
B
25](https://image.slidesharecdn.com/opencvtutorial-120522223338-phpapp01/85/Open-cv-tutorial-25-320.jpg)
![4. Distance Transform
1. Background
Video Model
Source Construction
2. Foreground Object 3. Upper body 4. Distance 5. Silhouette
Extraction Identification Transform extraction
DT [1] Poisson [3]
1 1 1 1 1
1 2 2 2 1
1 2 3 2 1
1 2 2 2 1 w(x, y) = 1 + ¼ * ( w(x+1, y) + w(x, y+1) + w(x-1, y) + w(x, y -1) )
1 1 1 1 1
27](https://image.slidesharecdn.com/opencvtutorial-120522223338-phpapp01/85/Open-cv-tutorial-27-320.jpg)
Open cv tutorial
- 1. Open CV Tutorial Ubiquitous Computing Lecture February 15th, 2012
- 3. Setup – Windows (1) • Download OpenCV 2.1 http://sourceforge.net/projects/opencvlibrary/files/opencv-win/2.1/
- 4. Setup – Windows (2) • Installation Add OpenCV DLLs into system path
- 5. Setup – Windows (3) • Include the header files and static libraries
- 6. Setup – Linux • Install OpenCV (say yes to any dependencies): $> sudo apt-get install libcv4 libcv-dev libhighgui4 libhighgui- dev libcvaux4 libcvaux-dev • To build: $> gcc -lcv -lhighgui –lcvaux YourFile.cpp
- 7. Links • Main wiki http://opencv.willowgarage.com/wiki/ • On-line library http://opencv.willowgarage.com/documentation/c/ • Tutorial http://www.cs.iit.edu/~agam/cs512/lect-notes/opencv-intro/opencv-intro.html
- 8. Getting Images from the Webcam Two Methods: • Event Driven (or interrupt driven) • New Frame == interrupt • Adv: constant FPS, multi-threaded processing • Dis: overhead, must be real-time • Polling (what we will use, the norm) • Grab frame when ready • Adv: only process what we can in real time • Dis: might grab same frame, variable FPS
- 9. Getting Images from the Webcam #include <stdio.h> #include "opencv/cv.h" #include "opencv/highgui.h" #include "opencv/cxcore.h" int main(int argc, char* argv[]) { CvCapture *capture = 0; //The camera IplImage* frame = 0; //The images you bring out of the camera capture = cvCaptureFromCAM( 0 ); //Open the camera if (!capture ){ printf("Could not connect to cameran" ); return 1; } cvNamedWindow( "demowin1", CV_WINDOW_AUTOSIZE ); //Create output window while(key != 27 /*escape key to quit*/) { //Query for the next frame frame = cvQueryFrame( capture ); if( !frame ) break; //show the raw image in the first window cvShowImage( "demowin1", frame ); //Get the last key that's been pressed for input key = cvWaitKey( 1 ); } //Shutdown cvReleaseCapture( &capture ); // WARNING: DO NOT release the "frame" variable cvDestroyWindow( "demowin1" ); return 0; }
- 10. Saving a video file int main(int argc, char* argv[]) { CvCapture *capture = 0; //The camera IplImage* frame = 0; //The images you bring out of the camera capture = cvCaptureFromCAM( 0 ); //Open the camera if (!capture ){ printf("Could not connect to cameran" ); return 1; } CvVideoWriter *writer; int isColor = 1; int fps = 25; // or 30 writer = cvCreateVideoWriter("output.avi", CV_FOURCC_DEFAULT, fps, cvSize(640, 480)); while(key != 27 /*escape key to quit*/) { //Query for the next frame frame = cvQueryFrame( capture ); if( !frame ) break; cvWriteFrame(writer, frame); //Just like showing an image //Get the last key that's been pressed for input key = cvWaitKey( 1 ); } cvReleaseVideoWriter(&writer); //Release the video file cvReleaseCapture( &capture ); // WARNING: DO NOT release the "frame" variable return 0; }
- 11. Accessing a pixel { int i = frame->height/2; int j = frame->width/2; //slow way CvScalar bgr = cvGet2D(frame, i, j); printf("B:%d, G:%d, R:%dn", (int)bgr.val[0], (int)bgr.val[1], (int)bgr.val[2]); //fast way (access the array directly) printf("B:%d, G:%d, R:%dn", ((uchar *)(frame->imageData + i*frame->widthStep))[j*frame->nChannels + 0], // B ((uchar *)(frame->imageData + i*frame->widthStep))[j*frame->nChannels + 1], // G ((uchar *)(frame->imageData + i*frame->widthStep))[j*frame->nChannels + 2]);// R //set the value cvSet2D(frame, i, j, CV_RGB(0,255,0)); cvShowImage( "demowin2", frame ); }
- 12. Transforming color images • Grayscale • 0.3*R+0.59*G+0.11*B, luminance IplImage* grayscaleImg = cvCreateImage(cvSize(640,480), 8/*depth*/, 1/*channels*/); { cvCvtColor(frame, grayscaleImg, CV_BGR2GRAY); cvShowImage( "demowin2", grayscaleImg ); } • HSV • What we perceive as color, rather than “sense” as color (well… sort of) • Hue: the color value • Saturation: the richness of the color relative to brightness • Value: the gray level intensity IplImage* singleChannelImg = cvCreateImage(cvSize(640,480), 8/*depth*/, 1/*channels*/); IplImage* hsvImg = cvCreateImage(cvSize(640,480), 8/*depth*/, 3/*channels*/); { cvCvtColor(frame, hsvImg, CV_BGR2HSV); cvSplit(hsvImg, singleChannelImg, NULL, NULL, NULL); // get the "Hue" component cvShowImage( "demowin3", singleChannelImg ); }
- 13. Reducing image size For reducing image size quickly (e.g., to increase frame rate) • Pyramid • nearest neighbor (no interpolation) • integer multiple down-sampling (throws rows and columns) • Anti-aliasing built in IplImage* smallerImg = cvCreateImage( cvSize(640/2,480/2), 8/*depth*/, 3/*channels*/); cvPyrDown( frame, smallerImg );
- 14. Image filtering For example low pass filtering cvSmooth(frame, frame, CV_GAUSSIAN, 15, 15); //size=15x15 CV_BLUR_NO_SCALE all ones (has ringing) CV_BLUR all ones, sum to one (has ringing) CV_GAUSSIAN a gaussian smoothing kernel (no ringing if large enough) CV_MEDIAN choose the median of the block (non-linear) CV_BILATERAL “perceptual closeness” (texture/color) of the pixels
- 15. Edge Detection The Canny edge detector //Requires a single-channel image: cvCvtColor(frame, grayscaleImg, CV_BGR2GRAY); //Reduce the image to the edges detected: cvCanny(grayscaleImg, edgesImg, hyst1, hyst2, sobelSize, l2needed?); 1. Image input 2. 3. 4. 5. Smoothing Gradient Local Hysteresis Maxima Thresholding Gx,y From magnitude hyst1 = image and direction of hyst2 = along lines Gx,y
- 16. Hough Circles //Requires a single-channel image: cvCvtColor(frame, grayscaleImg, CV_BGR2GRAY); //Want to smooth it to get less noise cvSmooth(grayscaleImg, grayscaleImg, CV_GAUSSIAN, 7, 9 ); //Detect the circles in the image CvSeq* circles = cvHoughCircles(grayscaleImg, storage, CV_HOUGH_GRADIENT, downsampling, minCenterDist, CannyThresh, AccumThresh minRadius, maxRadius ); • In General, Hough transform consists of 1. Edge detection 2. Accumulation in parameter space For each detected point, draw shape with different parameter Look for local maxima For a circle space is (x,y,R) R=5 R=10
- 17. Morphology Primer • Pick a structuring element • Cross • Disk • Rectangle • Pick an operation • Dilation (Max, “OR”) • Erosion (Min, “AND”) • Opening (E+D) • Closing (D+E)
- 18. Torso Tracking Ke-Yu Chen, 2012/02/15
- 19. Project goal • Track the human torso in real time from a single general- purpose camera 19
- 20. Motivation • High-tech camera is not available • Only have 2D sequences 20
- 21. Algorithm 1. Background Video Model Source Construction 2. 3. 4. 5. Silhouette Foreground Upper body Distance extraction Object Extraction Identification Transform 21
- 22. 1. Background Model Construction 1. Background Video Model Source Construction 2. Foreground Object 3. Upper body 4. Distance 5. Silhouette Extraction Identification Transform extraction … … 0~50 51~60 Background model 22
- 23. 2. Foreground Object Extraction 1. Background Video Model Source Construction 2. Foreground Object 3. Upper body 4. Distance 5. Silhouette Extraction Identification Transform extraction Fragmented pixels = - Shadow 23
- 24. 2-1. Fill the fragmented pixles 1. Background Video Model Source Construction 2. Foreground Object 3. Upper body 4. Distance 5. Silhouette Extraction Identification Transform extraction Closing by Reconstruction [5] 24
- 25. 2-2. Shadow detection [4] 1. Background Video Model Source Construction 2. Foreground Object 3. Upper body 4. Distance 5. Silhouette Extraction Identification Transform extraction R G Shadow pixel B 25
- 26. 3. Upper body identification 1. Background Video Model Source Construction 2. Foreground Object 3. Upper body 4. Distance 5. Silhouette Extraction Identification Transform extraction Th = 50% Th * HEIGHT Th = 60% Th = 70% 26
- 27. 4. Distance Transform 1. Background Video Model Source Construction 2. Foreground Object 3. Upper body 4. Distance 5. Silhouette Extraction Identification Transform extraction DT [1] Poisson [3] 1 1 1 1 1 1 2 2 2 1 1 2 3 2 1 1 2 2 2 1 w(x, y) = 1 + ¼ * ( w(x+1, y) + w(x, y+1) + w(x-1, y) + w(x, y -1) ) 1 1 1 1 1 27
- 28. 5. Silhouette Extraction 1. Background Video Model Source Construction 2. Foreground Object 3. Upper body 4. Distance 5. Silhouette Extraction Identification Transform extraction Threshold = 0.55 28
- 29. Smooth the result • Improve shaking results t-3 t-2 t-1 t + = Final Rectangular of the torso 0.4 0.6 29
- 31. Reference 1. Distance transform: http://en.wikipedia.org/wiki/Distance_transform 2. Siddiqi, K.; Member, S. & Kimia, B. B. Parts of Visual Form: Computational Aspects IEEE Transactions on Pattern Analysis and Machine Intelligence, 1995, 17, 239-251 3. Gorelick, L.; Galun, M.; Sharon, E.; Basri, R.; Society, I. C.; Society, I. C. & Br, A. Shape representation and classification using the poisson equation In In Proc. of CVPR’04, 2004, 61-67 4. Horprasert, T.; Harwood, D. & Davis, L. S. A statistical approach for real- time robust background subtraction and shadow detection 1999, 1-1 5. Closing by Reconstruction: http://artis.imag.fr/Members/Adrien.Bousseau/morphology/morphoma th.pdf 31
- 32. Future work • Dynamic threshold for shadow detection • Track multiple objects • Program to get ground truth (4 corners of the torso) • Adaptive morphological closing operation 32
- 33. Questions? 33