Multi-pedestrian Tracking Based on Social Forces
Hengle Ren, Fang Xu, Fengshan Zou, Kai Jia, Pei Di, Jie Kang · 2018
Multi-pedestrian tracking based on video has always faced many problems. Tracking-by-detection paradigm is a popular method to solve these problems. For example, due to the influence of sensors, lighting, background, detection may result in some false detections and missed detections. In order to solve this problem, in this paper, we propose a new tracking method based on the social force model. Here, pedestrians are divided into two categories: candidate pedestrians and real pedestrians. The real pedestrians are the pedestrians we want to track. Both can be transformed into each other by their respective historical records. The social force model is used to predict the position of each person in the next frame, and the weighted distance between the detected pedestrian in the current frame and the detection in the next frame of image is calculated. According to the distance matrix, the Hungarian algorithm is used to assign identities so as to achieve the purpose of multi-pedestrian tracking. Our results were evaluated on the MOT challenges dataset and compared with existing advanced algorithms. The results show that this method outperforms traditional algorithms in the number of mostly tracked (MT), mostly lost (ML) and the number of frames processed per second (FPS). Including Particle filter, traditional social force model and Kalman filter algorithm tracking method.