Aiming at the problems that deep learning models require considerable computational resources

leading to increased processing delays that are detrimental to real-time autonomous driving

a low-latency multi-vehicle cooperative perception system based on state estimation is proposed.This system realizes rapid response and precise understanding of complex traffic scenarios through highly efficient data fusion and state estimation.A state estimation pipeline is introduced

which significantly enhances the speed of system state update by optimizing data preprocessing

feature extraction and state update.A motion prediction method is designed to further improve accuracy and smoothness of perception.This method minimizes estimation errors caused by rapid movement of targets or occlusion while ensuring the continuity of perception.Experimental results demonstrate the superiority of the proposed approach over existing deep learning methods.On the same hardware platform

the proposed method achieves a computing latency of 37ms per iteration

which is 12.1 to 27.4 times faster than current deep learning methods which have latencies of approximately 599ms and 447ms.The positioning error is maintained within the range of 0.5~0.8m

fulfilling the system’s requirements for positioning accuracy.