Research And Experiment Of The Methods To Detect And Recognize Moving Targets In Video Surveillance

Intelligent video surveillance is a new direction and an attention frontier in the computer vision field, it is also the crystallization of high technology on computer science, machine vision, image project, pattern recognition, artificial intelligence, etc. The technology of moving target detection and recognition is key in Intelligent surveillance., which has wide application foreground in virtual reality, visual surveillance, perception interface, etc.Hardware frame and software architecture of the surveillance system designed in this thesis is introduced. The software is improved and the intelligence is enhanced , because the existing software has flaws about capturing images, recording images. While, for monitoring camera fixed in the environment, algorithms such as establishing background model, updating background model, removing shadow and recognizing targets are researched[0].Moving targets are detected by background subtraction and the establishing and updating background[0] method based on single-Gauss model is improved in this thesis. Foreground and background[0] is judged with the method based on statistics to effectively solve the problem of updating background. At the same time, by using the optical properties of shadows, the mathematical model of the shadow is proposed, with which.For upright human's recognition, height-width ratio, duty cycle and proportion which is between width of 6/7 object's height and object's height are used as the human characteristics. The improved algorithm of morphological contour extraction which extracting outline of the moving target is adopted by comparing commonly used methods. Then outline characteristic is extracted, the minimum distance classifier which based on linearly separable is designed. The models such as chair, ball, box and toy are used to verify result.The algorithms and methods of control are encapsulated in the different classes, the source codes are transplanted and modified easily, and the research result also can be seen as the good foundation of target tracking, behavior's analysis.