Crowd State Analysis JOHN BURNUM WORKING UNDER SALMAN
Crowd State Analysis JOHN BURNUM WORKING UNDER SALMAN KHOKHAR
Idea Analyze surveillance video of crowd motion to construct a scene containing: · predictions of dangerous regions due to crowd density · patterns of crowd motion to predict increased density in those regions
Current Method Calculate optical flow. Analyze to find patterns as in [1] and apply a pattern identifier as in [2] to find regions of convergent or divergent flow – they signify bottlenecks which could be dangerous. Using patterns, build a scene model which allows us to make some predictions about traffic flow. Mahdi has a paper which will provide an alternative method. [1] “Scene Understanding by Statistical Modeling of Motion Patterns. ” [2] “Similarity Invariant Classification of Events by KL Divergence Minimization. ”
Initial Results on Patterns from Optical Flow:
Results on Patterns from Trajectories:
Next Steps: Continue running current pattern codes with varying parameters to try for better results. Explore modifications to the algorithm or code to improve performance. Explore alternative options.
[1] “Scene Understanding by Statistical Modeling of Motion Patterns. ” Imran Saleemi, Lance Hartung, and Mubarak Shah. University of Central Florida and University of Wisconsin-Madison. [2] “Similarity Invariant Classification of Events by KL Divergence Minimization. ” Salman Khokhar, Imran Saleemi and Mubarak Shah. University of Central Florida.
- Slides: 7