Multi-Camera Trajectory Generation with Hierarchical Tracklet Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-camera tracking systems struggle to accurately generate trajectory information due to issues with intra-camera and inter-camera tracking, leading to incomplete and inaccurate integration of tracklets, especially when information is lacking.
Innovation Solution
A non-transitory computer-readable recording medium containing a trajectory generation program that performs preprocessing, global optimization processing, including ID transfer error removal, optimized hierarchical clustering, pose-oriented grouping, and error suppression to integrate and refine tracklets, ensuring accurate trajectory generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-camera tracking system uses conventional intra-camera and inter-camera tracking methods, then tracking process can be completed, but trajectory information generation accuracy deteriorates due to incomplete and inaccurate integration of tracklets
Solution Approach 1:
The patent segments the tracklet integration process into multiple hierarchical levels: first clustering tracklets into representative clusters based on appearance features, then further clustering into subclusters based on pose information. This multi-level segmentation allows systematic integration of tracklets while maintaining accuracy by addressing different aspects of similarity at different levels.
Solution Approach 2:
The patent adds another dimension to the clustering process by introducing pose-based clustering as a second level after appearance-based clustering. This dimensional extension from single-criterion to multi-criteria clustering enables more accurate discrimination between different persons and improves trajectory information generation accuracy.
2Measurement precision
If conventional tracking methods are used, then processing can be performed, but tracking accuracy deteriorates due to inability to accurately distinguish between different persons
Solution Approach 1:
The patent segments the clustering process into two distinct stages: first clustering based on appearance features to form representative clusters, then clustering based on pose features to form subclusters. This segmentation prevents mixing of different persons by addressing appearance similarity and pose characteristics separately, thereby maintaining person identification accuracy.
Solution Approach 2:
The patent applies different clustering criteria to different levels of the hierarchy: appearance-based clustering at the first level and pose-based clustering at the second level. This local differentiation of clustering quality allows precise distinction between persons with similar appearances but different poses, improving tracking accuracy.
3Ease of manufacture
If simple clustering is used, then processing is simple, but trajectory integration becomes inaccurate when information is lacking
Solution Approach 1:
The patent segments the clustering process into two hierarchical levels with different criteria. The first level uses appearance features for broad grouping, while the second level uses pose features for precise differentiation. This segmented approach maintains processing manageability while significantly improving integration accuracy, especially when information is incomplete.
Solution Approach 2:
The patent performs preliminary appearance-based clustering before pose-based clustering. This preliminary action organizes tracklets into representative clusters based on dominant appearance features, providing a structured foundation that simplifies subsequent pose-based differentiation and improves overall integration accuracy.
Data Source
AI summary
A non-transitory computer-readable recording medium has stored therein a trajectory generation program that causes a computer to execute a process including generating a fragmentary trajectory of the person based on the acquired analysis result calculating a feature amount based on a plurality of image frames associated with the generated fragmentary trajectory generating a plurality of representative clusters by clustering a plurality of specified fragmentary trajectories using a similarity of representative values of the calculated feature amounts of the plurality of image frames generating a plurality of subclusters obtained by clustering each of the plurality of clustered trajectories in each of the generated representative clusters using representative values of a plurality of poses of the person and setting the plurality of trajectories corresponding to the subcluster as a fragmentary trajectory of the same person based on a result of the plurality of generated subclusters.