多源数据融合的大型活动人流监测与分析技术

Crowd monitoring and analysis technology for large-scale events based on multi-source data fusion

为解决在举办大型活动时,仅利用单一数据源进行轨迹分析存在精准度不足、覆盖性不够和多场景功能受限等问题,以多源异构数据融合在大型活动人流监测与分析应用为目标,在系统设计和关键技术层面开展具体研究。在系统设计层面,提出多源异构数据融合整体框架,设计人流监测与分析系统方案;在关键技术层面,针对信令数据稀疏性、多样性特点,提出基于金字塔匹配的信令数据处理技术方案;针对跨场景视频融合分析难点,提出基于深度学习和时空关联的人员行为识别技术方案;针对人员轨迹分析精度不高难题,提出基于信令、视频、票证多源数据融合的轨迹还原及分析技术方案。在2022年北京冬奥和冬残奥场所应用验证该技术。结果表明:人群流向分布监测精度超过90%,定位精度小于100 m,实现人群轨迹的精准感知和预测。

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.