Human Surveillance System’s Design And Implementation Based On Adaptive Selection Of Features

Video surveillance is an important research direction in the field of computer vision. It has the widespread application prospect. The detection and recognition of moving human target is an important part of intelligent video surveillance system. It plays a vital role in the security of public places and sensitive sectors. So it is of great practical significance and worthy for scholars to further research the human body detection and recognition.In this paper, human body target surveillance system is designed and implemented based on adaptive features of human. Human target will show different features at different distances, different lighting conditions and under different sports attitude. If using fixed human feature for human detection, it will not reach the predetermined effect. Proper features should be chosen for detection in different situations, in order to improve the detection effect. In this paper, after the analysis of the brightness of the environment and human motion direction and distance, it adaptively chooses the most appropriate selection of the human motion features to detect. Meanwhile, in order to improve the practicality of the system, this paper increases the target zone alarm after the adaptively human detection, that is within the warning area will alert monitoring person if it detects an intruder was invaded. Experiments show that, the adaptive method of human motion detection area can be more reliable, more accurate.Human detection system in this paper is built on the Visual C++integrated development environment. It mainly detects and recognizes moving human target and gives alarms when it detects an intruder was invaded. In the adaptive feature selection part, this paper specifically analyzes the head shoulder features and gait features. Head shoulder model is obtained by horizontal and vertical projection histogram. After obtained the head and shoulder model, it was to extract Hu feature vector. Then we performed similarity comparison between the feature vector and the stand one which obtained through extensive experimental testing, according to the threshold to. judge whether the human body is. As for the gait feature, we use gait video in the database as the training sample, select eigenvalue statistics through the extraction of gait sequence profile, and then recognize human through the k nearest neighbor classifier.