Automobile detection based on feature tracking has proven effective in handling vehicle occlusions. However, current systems adopt a single camera to compute the 3D coordinates of feature points, which leads to an under-conditioned problem. As a result, only a few stable features can be reconstructed successfully. In this paper, we extend the system by adding a second camera, and integrate structure from motion (SfM) techniques to reconstruct a large volume of 3D points without sacrificing accuracy.Ninth IEEE International Symposium on Multimedia 2007 -Workshops 0-7695-3084-2/07 $25.00