Development Of Kalman Filter For Pedestrian Trajectory Prediction

Abstract

Computer vision is one of the main fields of artificial intelligence. Computer vision enables machines to extract useful information from images or videos. Applications include image recognition, autonomous driving, and pedestrian detection. In the context of traffic environments, the interaction between pedestrians and vehicles is a key focus of many studies. Since pedestrians often face more severe injuries in traffic accidents, therefore, accurately predicting pedestrian movement trajectories is crucial for improving traffic safety. To address this issue, combining object detection and tracking techniques to optimize pedestrian trajectory prediction has become a core direction of research. These technologies enable drivers or ADAS (Advanced Driver Assistance Systems) to have a more comprehensive understanding of the driving environment. This effectively reduces the risk of traffic accidents and enhances overall road safety. This study focuses on predicting the trajectories of pedestrians on the road. It calculates the pedestrian's relative speed and direction. It utilizes the Kalman filter algorithm to integrate object detection models, tracking models, and trajectory prediction models, by predicting the system state and correcting it based on new observed data. This continuously improves the accuracy of state estimation by using prediction and update steps, generating an optimal estimate through weighted averaging, in order to infer the pedestrian's potential future path. The research results show that the error between the x-axis data of the Kalman filter trajectory prediction and the real trajectory's x-axis data ranges under 0.1 %, while the error in the y-axis data ranges under 0.1 % in absolute value.