In this paper we described a new approach to background subtraction, which is lighting independent and suitable for real-time processing. We proposed the use of simplified stereo algorithm to perform the segmentation, which requires at least two camera views to be available. We demonstrated that the occlusion problem can be dealt with easily if more than two cameras are available.
Further work is required to refine the shown method of segmentation. The accuracy of the algorithm is related to the accuracy of the disparity map we obtain during the first, off-line, part of the algorithm. The precision can be increased if some non-linear interpolation of the triangulated mesh is performed instead of the linear one. We also plan to further investigate color calibrating the cameras during the off-line phase of the algorithm. This will help to better determine and reduce the required tolerance ranges for each camera pair.