Abstract:

Relying on only a single data source for trajectory analysis has problems such as insufficient accuracy, insufficient coverage, and limited functionality in multiple scenes when large-scale events are held. In view of the application of multi-source heterogeneous data fusion in crowd monitoring and analysis of large-scale events, system design and key technologies were studied in this paper. At the system design level, the overall framework of multi-source heterogeneous data fusion was proposed, and the crowd monitoring and analysis system scheme was designed. At the key technology level, according to the sparsity and diversity of signaling data, a technical scheme of signaling data processing based on pyramid matching was proposed. Due to the difficulties of cross-scene video fusion analysis, a technical scheme of human behavior recognition based on deep learning and spatio-temporal correlation was given. In view of the low accuracy of human trajectory analysis, a trajectory recovery and analysis technical scheme based on multi-source data fusion of signaling, video, and ticket was provided. The application of the technology in the 2022 Beijing Winter Olympics and Winter Paralympics shows that the monitoring accuracy of the crowd flow direction distribution is more than 90%, and the positioning accuracy is less than 100 m, which realizes the accurate perception and prediction of the crowd trajectory.

Key words: data fusion, large-scale event, crowd monitoring, multi-source heterogeneous, trajectory analysis