my scene is like this
i had set a camera in the back of the classroom .
what i want to do is when someone stand up i will know this .
for i am a beginner in opencv
i am not sure how to do this
especially the picture is seen from the back
is anyone who can give me some advice
thank you advance
best wishes
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)
I do encourage new people to try and solve problems, but please, do know what you are talking about. Pure segmentation of movement will fail in about 95% of the cases. Classes have windows, creating light changes, it will have a teacher running in the front, it will have a blackboard with changing information. I disagree on this approach.
I would suggest looking deeper into person detection using detection models. For example, the openCV interface supplies a Viola & Jones face detector, which has also a model for upperbody and full body detection. What I would suggest that you do is define a region where students are located, trying to exclude the teacher. Then apply the detection framework for the fullbody first, maybe the upperbody.
I am quite sure that people who are sitting down will not triger the model, but standing people will actually trigger it.
Give it a try, there are many topics out here on the subject of detection.
VJ doesn't perform very well on people detection, although it might be enough since we're talking about a controlled environment. However, I'd recommend using the HOG+SVM approach, since it has documented much better results in this task. It is a bit slower but I don't think that speed is much of an issue here.
@Median, actually, for the upper body detector it does quite good in those rather controlled conditions. Have multiple projects running with it, you just need to train a robuster model than the one contained in the sourcecode. As for the person detection, if it would be pure in the wild, I guess going with the DPM model is far better, just too bad it isn't included in its best form in openCV. About HOG+SVM I do agree that the results posted around are better, but it is a steep learning curve to get it to work decently. Documentation on it is terrible, the CPU part is far from documented and the CvSVM model is so outdated...
For the upper body detector, you do not need the face! It focusses more on contour properties, like head shape, body shape, arm transitions in gradient image, ... I would not be scared of that! The face model is just focussing on very specific details to actually grasp a face image location.
@StevenPuttemans thank you very much ! i tried you advice.the upper body detect can reach my scene .but it miss the target most the time .any way to make it much accurate ? sorry toboth you again !
Basically what you could do is make your own case specific upper body cascade classifier detector. Get yourself some training data and try it out?
Asked: 2014-01-17 01:09:28 -0600
Seen: 1,693 times
Last updated: Jan 17 '14
