Research on video pedestrian re-identification algorithm based on spatio-temporal dynamic information extraction

With the resurgence of the artificial intelligence research boom, pedestrian detection technology provides a brand new opportunity for intelligent processing and application of surveillance video. For the identification and tracking of pedestrians in surveillance videos, a pedestrian re-identification model based on spatio-temporal dynamic information extraction is proposed. The model contains two branches, namely the manual sequence feature extraction branch and the deep sequence feature extraction branch, which constructs a highly discriminative spatio-temporal feature representation for pedestrians in the video by adopting feature fusion techniques for different sequence features. Then, the pedestrian re-identification model is applied in practice to build an intelligent video surveillance system. Different datasets are selected for model ablation experiments and comparison experiments, and the results show that the models in this paper all outperform the baseline model, and the Rank-1 and mAP metrics are higher than the best classical algorithms by 4.28%~7.16% and 2.13%~4.71%, respectively. The combined recognition accuracy in the video surveillance system has improved by 5.60%, reflecting the model’s superior performance in pedestrian re-recognition.