For the problem of maneuvering target detection and tracking in low signal-to-noise environment, the improved track-before-detect algorithm based on mixture estimation of the multiple model particle filter is presented. Firstly, the current moment model is sampled by the previous moment particle model information and the model transfer probability. Then, the prediction for current particle is done by current measurement information, and the sequential importance resampling
smoothing way is presented, which can effectively increase particle diversity without increasing account task. Finally, the model and state are effectively estimated by the new particle muster. Simulation results show that the proposed algorithm has better performance of detection and tracking compared with the standard multi model particle filter.