BROWSE
Detail
Conference Paper
Simple Crowd Detection by Feature Map Reconstruction
Cited - time in 
Share 
Share
- Authors
- Sung-Uk Jung, Hyung-Keun Jee
- Issue Date
- 2015-11
- Citation
- International Conference on Computers, Communications and Systems (ICCCS) 2015, pp.1-4
- Language
- English
- Type
- Conference Paper
- Abstract
- Detecting crowd abnormality is an important step in terms of crowd monitoring, CRM system, and security. The previous researches usually depend on the dynamics of optical flow. Therefore, an object which does not move could not be considered for crowd analysis. To overcome the constraint, we propose a new method of detecting the crowd abnormality. We only focus on specific action such as crowd formation and evacuation. First, the extracted features of the crowd are clustered by K-Nearest Neighbor algorithm. Then, the moving features are filtered based on optical flow and the feature distribution is calculated. By the distribution of the crowd and the number of feature we reconstruct a new feature map. Lastly, the abnormality of the crowd is determined based on the reconstructed feature map and the distribution of the crowd. The usefulness and effectiveness is demonstrated by visual surveillance databases.
- KSP Keywords
- CRM system, Crowd Analysis, Crowd Monitoring, Feature map, K-Nearest Neighbor(KNN), K-Nearest neighbor algorithm, Optical flow, Visual surveillance, crowd detection, crowd formation, feature distribution