A Collision Detection Method Based on Vehicle Trajectory Data

A collision detection method based on vehicle trajectory data is proposed to solve the vehicle safety problem in vehicular networking. First, this method gives the algorithm to extract the outlier data segment from the denoised trajectory data based on time series containing acceleration sensor data. After formalizing the concept of collision according to domain knowledge, the outlier data segment is further transformed into the feature vectors of collision. Finally, a binary classification task on collision detection is constructed. In the experimental environment and practical application, this method can overcome the large noise of driving scene, and has a better precision and recall. In addition, the trajectory data are generated from inexpensive on-board electronic equipment, it can show strong versatility and feasibility in the production environment.