Flart's profile - activity

Gray is not so bright, as you suggested at comment. Try CV_THRESH_OTSU or cv::threshold(imgH, imgH, 150, 255, CV_THRESH_TRUNC), it's enough for that kind of images.

True, thank you. I have not seen it before. Usually "release" follows "release candidate".

Why OpenCV 3.0 gold? There are some "platinum" version coming soon?

May be there are some function in processing, what convert your image to 8UC3? Check docs to every function, what you are use, or use Image Watch Extensions for Visual Studio.

cv::threshold(img1, img2, 0, 255, CV_THRESH_OTSU);

Zero (threshold level) is ommited. How to get value, that algorithm uses ?

    int c = cv::waitKey(0);
    if (c == 2490368) y++; // up
    if (c == 2621440) y--;   // down
    if (c == 2424832) x--;   //  left
    if (c == 2555904)  x++;  // right

this is work for me. (dec)82 corresponds 'R'. http://en.wikipedia.org/wiki/ASCII May be that some locale issue? (which depend from system, too. Win, linux, etc.)

That value depends from architecture of your system. For example, for me "up_key" equal (int) 2490368.

Is there support read and write into animated gif in openCV?

Or really you want "de(spot)ing" your image?

If you blur image, you lose data. You can not restore it, so "deblurring" function don't exist (if I correctly understood this term)

You can try cv::normalize(), or addition contour to image for "improve" image

That simple code leads 20 error in xutility file, and when I had commented temp = distance(p, pi), then solution successfully built. How to fix code; and why did errors appear?

P.S. static analyzer tells me about type pi : cv::Point_<int> pi, is that relates to the problem?

typedef vector<cv::Point> Points;

double distance (cv::Point& p1, cv::Point& p2) 
{
    return sqrt( pow( p1.x - p2.x,2) + pow( p1.y - p2.y,2));
}

double distance (cv::Point& p, Points& ps) 
{
     double d = numeric_limits<double>::max();
     double temp;

     for (auto pi: ps) 
     {
          temp = distance(p, pi);
          if (temp < d) d = temp;
     }
     return d;
}

nice question, waiting big_karma_guys )

I'm using HOG based pedestrian detector, but reduction false-positive leads increasing false-negative, so there are many skip pedestrians. So I can't determine _direction_ only by hog detector out_rectangles, because there are several pedectrians in one frame. And I using these out_rectangle with calcGlobalOrientation (see code above), but it don't work correctly. ----- updated my question.

I'm using HOG-based pedestrian detector, but reduction false-positive leads increasing false-negative, so there are many skip pedestrians. So I can't determine _direction_ only by hog detector out_rectangles, because there are several pedectrians in one frame.

So, I'm trying detect moving direction of found pedestrian by different way. One of a way is using motion gradient ( these out_rectangle to calcGlobalOrientation (see code below)), but I have unacceptably wrong result. I don't understand, why it are not handled correctly (upperbody doesn't have some changing area).

code:
// ... HOG-detector
Rect r_hog // result of HOG

cvtColor(frame, frame, CV_BGR2GRAY );
cv::absdiff(frame, preframe, diff);
cv::threshold(diff,diff,30,255, THRESH_BINARY);

cv::morphologyEx( diff, diff, cv::MORPH_CLOSE, cv::Mat( 5, 5, CV_8UC1 ) );
cv::morphologyEx( diff, diff, cv::MORPH_OPEN,   cv::Mat( 5, 5, CV_8UC1 ), cv::Point( -1, -1 ), 1, cv::BORDER_CONSTANT, cv::Scalar( 0 ) );

Size imageSize = diff.size();
Mat m_motionHistoryImage( imageSize, CV_32FC1,0.0 );
Mat m_segmask( imageSize, CV_32FC1 );
Mat mask ( imageSize, CV_8UC1 );
Mat orientation( imageSize, CV_32FC1 );

double cycleTime = 0.04;
double m_maxMotionGradient( 1.5 * cycleTime );
double m_minMotionGradient = m_maxMotionGradient/10;
double m_motionHistoryDuration(7 * cycleTime ) ;
double time=clock();

double timestamp = (time - startTime) / CLOCKS_PER_SEC/1000.0;
const double min_conrour_area = 200;

cv::updateMotionHistory( diff, m_motionHistoryImage, timestamp, m_motionHistoryDuration );   
calcMotionGradient( m_motionHistoryImage, mask, orientation, m_minMotionGradient, m_maxMotionGradient, 3 );

Rect r_without_legs = Rect(r_hog.x,r_hog.y, r_hog.width, r_hog.height/2);

double angle = calcGlobalOrientation( orientation(r_without_legs), mask(r_without_legs), m_motionHistoryImage(r_without_legs), timestamp, m_motionHistoryDuration);
angle = 360 - angle;

// draw angle

Thank you, I have missed that cv::getTextSize function :)

I want to write value on picture using puttext with black background. Code:

rectangle(frame, Rect(
     Point(50,50) + Point(0,-height), 
     Point(50,50) + Point(width,0)), 
     Scalar(0,0,0),-1);
putText(frame, Tostr(value), Point(50,50),1,1,Scalar(255,255,255),2);

But change of capacity value leads change of required width. How this width is determined, or did I miss some params of putText, which resolve my issue?

If chinayin _really_ beginner then he can start with my sketch, otherwise you are right, of course. Thank you for encouraging :)

sorry, "true height" -- it is man's height at the picture by pixels. I forgot about IRL world and what "true height" is it by meters :-)

I notice, then I increase HeightPedestrian from real 250 pixels (measured by A Ruler For Windows http://www.arulerforwindows.com/ from my video, which on screen thanks to cv::imshow) to 350 -- quality of detector increase. So I propose that in resizeRatio term actually not 128.0, but lower. Are you sure what all pedestrians were resized towards a 64x128 window from top to heel, without sky and ground on top and bottom of image?

Put your camera at level of heads (when people sit), consider part of images (video) above heads, and then use algorithm, base on frame difference.

When every one is sit -- where is no change in area above heads, when someone stand up -- he create big difference, that difference is your classifier (everyone sit / someone stand up)

HOGDescriptor::getDefaultPeopleDetector() classifier was trained using training images of 64*128 size, but what real height pedectrians at those train data?

Performance is important, and I approximately know height of my pedestrians -- so I using HOGDescriptor::detect() (not detectMultiScale()).

Firstly I resize my target area to detector resolution (coefficient of resizing equal to ratio between "detector height pedectrian" and "my data height pedectrian"), and then perform detect function.

code:

double HeightPedestrian = 250.0;
double resizeRatio = 128.0 / HeightPedestrian;
resize(frame(RectWindow), frame_for_HOG, Size(), resizeRatio, resizeRatio);
HOGDescriptor hog;
hog.setSVMDetector(cv::HOGDescriptor::getDefaultPeopleDetector());
hog.detect(frame_for_HOG, foundPoints, 0, Size(8,8), Size(32,32));