Agreed, but my findings are that the false detection of background movement is exacerbated if the camera pans across the scene. I would have thought that if the camera scans a scene in a predictable manner they could just XOR two aligned scenes pixels to see what moved a computationally simple exercise. IMO leaf movement will provide only minor errors along the edges of the leaf from one scene to another and can be discounted in an algorithm. The system only needs to look for large blocks of pixels movements within the scene to determine that there is something new here. This may be one way to tackle person detection within a pan scan as currently It only seems to work in my case with the pan scan disabled and employing the pan scan as a stationary camera. Altr=ernatives to the above would be to calculate a correlation function value for each matched scene and only look deeper if the match error exceeds a user-settable threshold.