This paper deals with the temporal analysis of stereo image sequences taken from a road vehicle in a busy traffic environment. The images are first processed to extract contours of significant intensity change. Points lying on these contours are used as primitives for stereo matching and optic flow computation, which yield the 3-D trajectories of the salient points in the scene. These point trajectories are processed by a Kalman filter to determine the relative motion of the points with respect to the vehicle. The 3-D output of the stereo algorithm is grouped by a segmentation algorithm into different objects. A statistical analysis of the motion parameters is performed for each object. Experimental results with a real stereo sequence are presented.