Description:
- Security surveillance
- Driving assistance systems
- Smart home
Abstract
USC researchers introduce a three-level hierarchical association approach to track multiple objects in crowded environments using a single camera. At the low level, reliable tracklets are generated by linking detection responses with conservative affinity constraints. These tracklets are then further associated at the middle level based on more complex affinity measures. The high level estimates entries, exits, and scene occluders using the computed tracklets, refining the final trajectories. Applied to pedestrian tracking, our approach outperforms previous methods, demonstrating significant performance improvement. This innovative solution enhances object tracking accuracy without significant increases in computational cost.
Benefit
- Improved tracking accuracy
- Flexible associations
- Efficient computational cost
- Reliable tracklet generation
Market Application
Tracking methods for computer vision often suffer in complex and and crowded environments. Existing feature-based tracking methods lack a discriminative model to distinguish the object category of interest from others, leading to missed detections and inaccurate responses. Despite advancements in object detection techniques, accuracy remains imperfect.
Publications
Robust Object Tracking by Hierarchical Association of Detection Responses, Huang et al., 2008.
Other
Stage of Development
- Experimentally validated on challenging pedestrian datasets
- Available for licensing