Related Experiment Videos
A novel method for tracking pedestrians from real-time video.
Jian-qiang Huang1, Xiang-xian Chen, Le-yu Wang
1Department of Instrumentation Science and Engineering, Zhejiang University, Hangzhou 310027, China. abraham_hjq@yahoo.com
Journal of Zhejiang University. Science
|December 10, 2003
Summary
This study introduces Pedestrian Tracking using Support Vector (PTSV), a novel method enhancing video surveillance. PTSV improves tracking reliability by using Support Vector Machine (SVM) classification scores instead of traditional optical flow intensity differences.
Area of Science:
- Computer Vision
- Machine Learning
- Video Surveillance
Background:
- Traditional optical flow tracking relies on minimizing intensity differences between frames.
- This method can be unreliable with large object motions or appearance changes.
Purpose of the Study:
- To introduce a novel Pedestrian Tracking using Support Vector (PTSV) method for video surveillance.
- To improve tracking reliability compared to traditional optical flow methods.
Main Methods:
- Integrating a Support Vector Machine (SVM) classifier into an optic-flow based tracker.
- Tracking objects by maximizing SVM classification scores, not minimizing intensity differences.
- Utilizing image pyramids and a coarse-to-fine scan for large motion handling.
- Employing the Sequential Minimal Optimization (SMO) method for accelerated SVM training.
Main Results:
- PTSV demonstrated improved tracking reliability in real-time video surveillance.
- Comparative experiments confirmed PTSV's superiority over traditional optical flow tracking.
- The method effectively handles object classification without direct pixel region comparison between frames.
Conclusions:
- PTSV offers a more reliable approach to pedestrian tracking in video surveillance.
- The integration of SVM classification significantly enhances tracking performance.
- This method provides a robust solution for challenging tracking scenarios.