Driver’s dangerous driving behavior detection method based on trajectory data

Studying methods for detecting dangerous driving behaviors is of great practical significance for improving road safety and reducing accident rates. Therefore, a driver's dangerous driving behavior detection method based on trajectory data is proposed. To address the issue of class imbalance in vehicle GPS trajectory data, an adaptive synthetic sampling method is adopted for data balancing processing, A trajectory data reconstruction model based on an encoder decoder architecture was constructed, which utilizes self attention and cross attention mechanisms to deeply integrate vehicle trajectory features with road map information to output data reconstruction results, Construct a semi supervised projection twin vector machine model that distinguishes dangerous driving behavior based on the reconstructed trajectory data sample projection and class center distance. The experimental results show that the average false positive rate of the proposed method is 3.41%, the average false negative rate is 3.14%, and the minimum detection time is 0.24s.