Research Of Intelligent Video Surveillance Technology In Family Environment

Intelligent Video Surveillance means that it can automatically analyze the videos from video camera by the method of computer vision and video analysis and judge the deeds by people in the video scene without artificial intervention, thus the various deeds can be intelligently disposed or alarmed accordingly. This topic is to research and develop the intelligent surveillance system on family environment. The system is based on movement target test and tracking module, defines and identifies the deeds by people in the video, it can use on the intelligent electrical appliance control, human-computer interaction for later family and office. The human action recognition research on the computer has been carried in domestic and overseas for many years, this paper mainly focus on researching two category of action recognition method aimed at the action particularity on the environment of family, namely static image method based on tracking and key point analysis method based on time and space. In addition, there are special video data for experiment. The following is the work done by this paper:1. To study and compare various of background modeling method, aimed at the feature that many of home video surveillance system applied to dynamic scenes. adopt mixed Gauss method and multi-information fusion background modeling method to automatically remove noises and test foreground.2. For target tracking, a major difficult point is that the shielding, merger, separation between moving targets in the process of tracking. To this point, a tracking method put forward by us that adopt feature matching method that based on color histogram similarity to match the objects on continuous image.3. To put forward a special static image recognition method aimed at three category movements including stand, sit down and lie down on family environment. It mainly uses drawing ellipse and body clothing color information to extract feature, and set the parameter threshold formula to complete final recognition process.4. To research a variety of feature description methods based on interest point, to adopt an improved SVM method for the final identification, and photographed six kinds of action video database on family environment. The experiment results show that the proposed two kinds of action recognition method have a good recognition results to the particular action on the family environment.