Study On Image Processing And Recognition Techniques For Intelligent Video Monitoring System

Intelligent video monitoring system has wide application in many scopes. It is the important high-tech method and technology to guarantee the public security of modern society. Traditional video monitoring system cannot achieve the goal of unmanned safeguarding due to lack of automatic recognition performance. Aiming at the drawback of traditional video monitoring system, this paper mainly concentrate on the research of image processing and recognition techniques for intelligent video monitoring system. Based on this goal, the following research achievements have been fulfilled:(1) Moving object detectionThe detection of intruding moving object is one of the key techniques for intelligent video monitoring system. In orcer to extract the moving object area robustly in complex outdoor background, a moving object detection method based on a color image difference model, an adapti\e thresholding method, and an image morphology filtering method was proposed Moreover, several image smoothing, enhancement and edge detection methods were analyzed and compared. Experimental results demonstrate that the proposed approach can detect moving objects effectively in outdoor surveillance environments.(2) Shadow eliminating for moving objectIf the shadows produced by change of illumination exist, the moving shadows will be misclassified into parts of these objects. This will cause consequent recognition process of the objects to fail. Therefore, based on histogram and clustering techniques an efficient multi-object shadows distinguishing and eliminating method for outdoor surveillance environmeit was developed in this dissertation. Moreover, C-means clustering method anc fuzzy clustering method were also analyzed and compared. The results of experiment show that the shadow distinguishing and eliminating approach for malti-object is fast and effective.(3) Moving object recognitionThe change of detected object region made by movement brings difficulty for object recognition correctly. For the solution to this problem, two recognition approaches were proposed in this paper. One is the BP neural network recognition method based on moving object's whole shape characteristics and the other is GA neural network recognition method based on moving object's head-shoulder shape characteristics. The former first eliminate the shadow of moving objects, and then extract shape invariant moment characteristics of object's whole region and identify them by a BP neural network. The latter directly extract the moving object's head-shoulder shape characteristics and then identify the object by a GA based neural network without the shadow distinguishing and eliminating operations. So this method can identify object rapidly and correctly.