Revision history [back]

I would try to detect persons using the person model of latent svm: http://docs.opencv.org/modules/objdetect/doc/latent_svm.html

or using HOG: http://docs.opencv.org/modules/gpu/doc/object_detection.html

Then, in each frame, I would locate the person in the frame, save the bounding box around it from frame to frame and check if the bounding box changes. If the location/size of the bounding box changes, then a motion is detected.