Automotive radar can be used to detect pedestrians and vehicles and keep stable tracking of the targets. Multi-targets tracking is the key techniques when tracking in the complicated road condition. Some targets may lose alarm and there may be some false targets among the measurement because the radar would be affected seriously in the complicated road condition especially by the clutter and multi-path effect. Tracking can solve the effect of the false targets to a certain extent and provide a stable and accurate state of the targets. How to associate the track and the measurement is important in multi-targets tracking system. A robust tracking algorithm using joint integrated probabilistic data association is proposed. Unlike the nearest neighbor method, all the possible combinations of track measurement assignments are considered and the probabilities of the joint events are calculated. The probabilities of the individual track are calculated recursively which allow us confirm and delete the tracks. Meanwhile targets may maneuver when the vehicles change the lanes in the fast-changing road conditions. Traditional method has a low tracking precision because it assumes the targets motion are known and the predict state is calculated using the hypothetical state equation. Interactive multi-model filter is used to solve the problem which assumes the motion of the target may be interactive of the multi-models and weights the result of each model. We simulate some typical road condition to show the effectiveness of the algorithm comparing to other algorithm, and verify the algorithm using the real data collecting from millimeter-wave Radar.