Trajectory Extraction and Optimization Method After Emergency Video Multi-target Tracking

CHEN Siyuan; HU Xiaowu; ZHANG Chen; JIANG Wenying

doi:10.3969/j.issn.1672-9722.2025.05.015

Abstract：The trajectory of people in emergency scenes has strong research value,and it has guiding significance for people's emergency avoidance in similar scenes in reality.However,due to the sudden nature of the event,it is difficult to obtain pedestrian trajectories in this scene through traditional GPS or sensors.With the wide application of surveillance technology and continuous breakthroughs in the field of computer vision,it has become a good solution to extract pedestrian trajectories from emergency videos through multi-target tracking algorithms.In this paper,the FairMOT multi-target tracking algorithm is used to extract and prepro-cess pedestrian trajectory data in emergency videos.Compared with the traditional two-stage algorithm,the multi-target tracking al-gorithm used has a faster inference speed.The traditional one-stage approach has fewer ID switches.In the actual extracted trajecto-ry data,this paper finds through simulation experiments that the trajectory point data extracted in this way has a lot of redundancy and the phenomenon of trajectory point jitter.In response to these problems,this paper proposes a trajectory optimization method that combines trajectory point extraction and Savitzky-Golay(S-G)filtering,and selects five different scene emergency videos for experiments.Experiments show that the optimized trajectory obtained by the method in this paper can effectively alleviate the phe-nomenon of trajectory point redundancy and trajectory point jitter,retain the real trajectory characteristics while reducing the amount of data,and can further provide for the planning of pedestrian trajectory data analysis and personnel evacuation in emergen-cy scenarios data support.