We introduce a realtime vehicle detection and tracking algorithm for in-vehicle video images. Although various vehicle detection approaches have been proposed, it is difficult to find a fast and reliable algorithm for realtime applications, such as for vehicle collision warning. We introduce a realtime appearance-based vehicle detection approach. It uses constrained Delaunay triangulation to make image evidence collection for multiple overlapping hypotheses computationally efficient. We also propose an integrated detection and tracking approach where redundant detection and tracking can compensate each other's shortcomings. The experiment was performed on various video clips of highways and local roads with various traffic and illumination conditions, and the resulting performance is competitive compared to those of other active sensors