Vehicle tracking trajectory optimization algorithm based on bidirectional long short-term memory with Savitzky–Golay in a vehicle-road cooperative environment
Zijian Wang, Xin Cheng, Xinpeng Yao, Zhou Zhou, Han Zhang, Hui Chang · Journal of Intelligent Transportation Systems · 2025
This article proposes a vehicle tracking trajectory optimization algorithm for use in a vehicle-road cooperative environment, designed to solve the current problems of low vehicle tracking accuracy, object jump, and missing and unsmooth trajectory data. The BiLSTM-SG algorithm developed in this study leverages multiple methods to optimize trajectories of moving vehicles. Considering the temporal sequence of position changes in moving vehicles, the LSTM object frame prediction algorithm incorporating vehicle speed on object frame movement is used to improve the DeepSORT tracker. To make full use of the temporal sequence of a moving vehicle’s trajectory and contextual information, the BiLSTM trajectory bidirectional completion algorithm is used to complete the missing segments of vehicle tracking trajectory and eliminate error accumulation. And then the completed trajectories are optimized according to the Savitzky–Golay (SG) trajectory smoothing algorithm. The algorithm is trained and tested on the vehicle detection and tracking data set UA-DETRAC. The experimental results show that the algorithm effectively reduces the jump rate of the vehicle tracking object, improves the accuracy, and robustness of vehicle tracking, eliminates the error accumulation problem to a certain extent. It provides important data support for traffic information processing and analysis in the vehicle-road cooperation environment.