Collaborative Surveillance And Real-Time Abnormal Detection In Large-Scale Camera/Sensor Network
- French Team Leader: Hichem Snoussi — LM2S, UTT
- Chinese Team Leader: Ma Shiwei, SMEA/Automation Department and Wang XiangYang, School of Communication & Information Engineering, SHU
- Team: Hichem Snoussi, Shiwei Ma, Wang XiangYang, Zhu QiuYu, Chen JunLi
Smart public safety system is an important part of the Smart City. The large-scale video surveillance systems, which play a vital role in improving management of urban safety and maintenance of social stability, have become infrastructure for social management. This project focuses on two use of a smart video surveillance camera network-based system: collaborative image processing in camera networks and video stream-based abnormal detection.
2. Collaborative image processing in camera networks
Objective of multi-camera pedestrian tracking is to use different camera to overcome obstacle and achieve the specified target tracking. The first challenge is the pedestrian feature extraction. The second challenge is feature matching in different camera. We focus on robust pedestrian feature des- criptor and feature matching method which includes non linear classification, Bayesian filtering and manifold learning.
3. Video stream-based abnormal detection
For prevention of public safety incident, we need real-time detection of abnormal behavior of indivi- duals and groups based on the video data. The first challenge is modelize pedestrians and infer from it individual behavior. The second challenge is how to define what is an abnormal behavior. The third challenge is to understand group behavior. Here we focus on the pose estimation based on the depth information and skeleton, semi-supervised Boosting learning algorithm for tracking determines and support vector machine algorithm for abnormal group behavior detection.