Research And Realization Of Key Technology Of Behavior Recognition In Intelligent Monitoring System

Intelligent monitoring system reflects the future trend of digitization of videosurveillance and intelligent development. It has a wide range of potential application inthe field of transportation, military, public safety and has become the popular issue inthe field of computer vision. This thesis is based on the research and analysis of therelated technology, carrying out the experimental research on the main technology ofthe recognition of abnormal human behavior at the static background.In terms of the moving object detection, this thesis proposes an improvedalgorithm of the detection of moving object based on Gaussian mixture model algorithmfor background, detecting the moving area by means of three frame differencemethod.After using the area-method to determine the emerge of moving object, and it isdetected and extracted by the improved Gaussian mixture model algorithm forbackground. On the condition of meeting the requirement of instantaneity, it ensures agood effect of object detection.In terms of the tracking of moving object, this thesis researches on Mean-shifttracking method and the characteristics of the SIFT feature points and puts forward theSIFT feature points modulus value-direction histogram instead of the color features ofMean-Shift tracking algorithm. The proposed algorithm overcomes the shortcomingspossessed by the original one that when the color of object closes to the background orthe object is sheltered, and the failure detects are appeared.In terms of the recognition of abnormal behavior, this thesis pedestrians thecharacteristics of abnormal behavior such as jumping, hovering, run and illegal invasion,etc., and it defines detection rules of these types of abnormal behavior and proposes thecorresponding abnormal behavior detection algorithm. This method detects theabnormal behavior of jumping and squatting walk by means of extracting the trajectoryof a moving object and the aspect ratio of the external rectangular box. It detects thewandering of pedestrians by analyzing and counting the changes of the coordinates ofthe center of mass of moving object. According to the coordinates of the center of massof the moving object and the relationship between it and the warning area, it can judge whether the pedestrians break into the warning area or not. Besides, it is able tocalculate the rate of movement of moving target and detect the running of pedestriansaccording to the changes of object centroid with the time.This thesis takes simulation experiments for each of these algorithms, and theresults show that the proposed algorithm can quickly and efficiently detect abnormalbehavior of the traveling people.