A novel multi-sensor data fusion based multi-vehicle tracking algorithm and a longitudinal collision warning strategy were proposed, in order to achieve the safety-enhanced multi-vehicle tracking in dense clutter environments. The time registration and spatial fusion approaches were introduced, with respect to the problems of time asynchronous sampling sequences of multi-source sensors caused by nonidentity of sampling periods, sampling start-up time and inherent communication time delay, as well as different coordinate systems and different space dimensions. The single sensor multi-object state estimation based on the improved joint probability data association (JPDA) was utilized to estimate the multi-object trajectory, which not only ensured the effective association, but also could reduce the computational complexity to a certain extent. Furthermore, the multi-sensor joint probability data association (MSJPDA) sequential filtering algorithm was employed to update the motion state of targets serially, and the output of the last sensor was utilized as the state estimation of the fusion center, by which the situation of longitudinal collision could be well-estimated. Results of comparative experiment demonstrated the feasibility and effectiveness of the tracking algorithm proposed.