Anomaly Detection In Surveillance Scenes

The public security problem has attracted much attention. Nowadays,monitoring equipment has been installed on every corner. Traditionalsurveillance system which relies on manually supervising needs a lot oflabor. Intelligent monitoring technology plays an important part in smartcity development.In various issues of video surveillance, anomaly tends to be the mostimportant problem, which can be interpreted as different behaviors indifferent scenarios. Anomaly analysis involves target detecting andtracking, atomic motion recognition and activity analysis, group behaviorunderstanding, and many other video surveillance technology. Our study ismainly devoted to analysis anomaly in surveillance scenes.We propose two different approaches for anomaly detection to handledifferent scenarios. One approach is based on target detecting and tracking.The other one relies on texture information. The first approach achieveshigh detecting rate on scenes in which moving target could be exactlylocated. The second approach can be applied on crowded scenes. Moreover,we take attention on two special applications: violence detecting onsurveillance scenes and elderly care issue. To detect violence in videos, weput forward the idea to use optical flow-Gaussian model approach fordetecting "big movement" from the image. Then we use multi-scalescanning scheme to search violence region and realize a quick detectingmethod. To handle elderly care issues, we propose an approach combinedwith R-transform and Hidden Markov Model method. First, backgroundmodel is used for contour exacting. Then we use R-transform to describebody’s posture. At last, hidden Markov model is applied for gesture state analysis and realizes abnormal behaviors detecting in daily life.