Object Tracking via Similar Group Segmentation

The patent segments the tracking process into two distinct phases: (1) candidate generation phase that uses non-spatio-temporal similarity to identify potential matching objects across frames, and (2) verification phase that applies spatio-temporal constraint checks to validate candidates. This segmentation allows each phase to optimize for its specific purpose without compromising the other.

The patent introduces spatio-temporal constraints as an intermediary mechanism that mediates between the flexible non-spatio-temporal similarity matching and the requirement for accurate tracking. The constraints act as a filter that refines the candidate objects generated by similarity measures, ensuring they also satisfy temporal and spatial consistency requirements.

The patent implements periodic batch processing at predetermined intervals to assign tracking IDs to similar object groups. This periodic action consolidates multiple tracking decisions into efficient batch operations, improving overall throughput while maintaining acceptable response times through the use of intermediate storage for accumulating objects awaiting batch processing.

The patent performs preliminary object grouping and similarity calculation before the actual tracking ID assignment. Objects are pre-processed and organized into similar object groups based on feature similarity, so that when batch processing occurs, the computational workload is already reduced and objects are ready for efficient ID assignment.