Robust denoising and moving shadows detection in traffic scenes

This paper presents a work we have done on the motion detection in the context of an outdoor traffic scene for visual surveillance purposes. Our motion detection algorithm is based both on background subtraction and three frame difference. We propose quite innovative solutions for denoising, blobs filling and shadow detection without exploiting any a priori knowledge. Actually, methods presented here have been fully setup only for the former technique. Target sequence is made of 8-bits grey level still images, taken at 30 fps. This application works off line at 8 fps on a 800 MHz Pentium III computer.