Figure 4 Pedestrian re-identification process and results
Posted: Thu Jan 23, 2025 9:23 am
The pedestrian trajectory tracking service uses computer vision models to intelligently analyze surveillance videos and realizes automated personnel search through three main functional blocks: personnel identification, tracking, and retrieval. Pedestrian re-identification is the core capability of the entire system. It is "a technology that uses computer vision technology to determine whether there are specific pedestrians in images or video sequences", enabling the system to automatically screen specific personnel in surveillance videos through "image search".
In the system construction process, there is first the challenge of heterogeneous data across cameras. The camera shooting scenes are complex, including lighting changes, severe occlusion, and a large number of tracking targets. Single-point target tracking algorithms cannot achieve good results; there is also the challenge of limited side resources.
KubeEdge Sedna combines the multi-point computing power kuwait mobile phone number list of edge nodes to fully utilize the resources of edge nodes; supports collaborative reasoning based on feature extraction to protect data privacy; supports filtering of edge data to reduce the amount of data transmitted at the edge. At the development level, KubeEdge Sedna provides detailed application API templates and edge intelligence cases. KubeEdge Ianvs carries benchmarks for related processes, data, and algorithms, providing an important foundation and solution reference for this pedestrian Re-ID AI system.
Its support for mainstream AI tools also greatly reduces the cost of AI algorithm development and deployment. Combined with its own high reliability and large-scale node management capabilities, it meets the actual business needs of pedestrian trajectory tracking. It has been applied in BJXA Park and SZZC agencies in Beijing, Guangdong and other regions. During the application process, the multi-target tracking accuracy (MOTA) increased by an average of 23.22%, and the inference latency decreased by an average of 12.74%.
In the system construction process, there is first the challenge of heterogeneous data across cameras. The camera shooting scenes are complex, including lighting changes, severe occlusion, and a large number of tracking targets. Single-point target tracking algorithms cannot achieve good results; there is also the challenge of limited side resources.
KubeEdge Sedna combines the multi-point computing power kuwait mobile phone number list of edge nodes to fully utilize the resources of edge nodes; supports collaborative reasoning based on feature extraction to protect data privacy; supports filtering of edge data to reduce the amount of data transmitted at the edge. At the development level, KubeEdge Sedna provides detailed application API templates and edge intelligence cases. KubeEdge Ianvs carries benchmarks for related processes, data, and algorithms, providing an important foundation and solution reference for this pedestrian Re-ID AI system.
Its support for mainstream AI tools also greatly reduces the cost of AI algorithm development and deployment. Combined with its own high reliability and large-scale node management capabilities, it meets the actual business needs of pedestrian trajectory tracking. It has been applied in BJXA Park and SZZC agencies in Beijing, Guangdong and other regions. During the application process, the multi-target tracking accuracy (MOTA) increased by an average of 23.22%, and the inference latency decreased by an average of 12.74%.