检测STATIC摄像机图像中的人物的最佳方法（不使用视频流）-OpenCV中的HOG，Haar，BackgroundSubtract？

Question

So I am planning on implementing human detection static camera images - the camera never moves but the lighting conditions do change. Humans can be upright or could be sitting.

Question:

What would be better in this case to detect people in such setup?

Haarcascade mostly deals with faces and eyes and my images are taken from an angle where these features are extremely hard to detect especially for people with their backs toward the camera.

HOG features - I am not sure on this one

backgroundSubtractorMOG - does not seem to provide a very good result now even using a lot of images to "run through the subtractor" to build the background

Sample image with no people in it:

Answer 1

Maybe it would be easier to detect anything that moves on a picture? Detecting body, that can be in different positions would be very difficult (especially if half of the body is hidden behind a table). I'm writing face detection algorithm using Haarcascade at the moment - it can deal with any type of shape (not only face and eyes), but it must find some common features between those objects - bodies can vary a lot in colours, shapes and positions. It would also be difficult to make a cascade for that (you'll need thousands of sample images)

If camera is static, the only things that move would be humans. Utilize that fact. I don't have experience in that matter, but this place may give you a hint: http://www.codeproject.com/Articles/10248/Motion-Detection-Algorithms