Ask Your Question

Revision history [back]

click to hide/show revision 1
initial version

about traincascade negative set creation


I'm training a classifier with opencv_traincascade.exe for a VideoSurveillance AirCam positioned in a fix place outside one office. I created about 1000 positive samples. Reading many tutorials I find out that the number of negatives should be more than the positive, BUT I doubt it in this case as the camera is fixed in a still place and the background is always the same. So my concerns is to create a dataset of negative images that is only about one hundred images, cropping different details of the same, and never changing background. Is it correct? Can anybody give me any hits. A suggestion will be welcome, and helpful. Thanks.