Moving Human Object Tracking And Abnormal Behavior Recognition

As the developing of scientific and technology and the growing of intensity of population, the scienty is becoming more and more complex, abnormal accidents are increasing too, which can do harm to the benefits of the country and people. Tracking moving human object and abnormal behavior recognition is an important area of the area of research in pattern recognition and computer vision technology, simultaneously it's also the key to intelligent monitoring technology's widely use. So in this paper we do a depth research focused on the tracking moving human object and abnormal behavior recognition based on monocular video and we prove the validity of the method by the experiments through using it into the moving human database.In the phase of tracking moving human object and abnormal behavior recognition, the moving human object detection is the key foundation. This paper firstly introduces the basic method of moving human detection which includes optical flow method, frame differencial method, background subtraction method, and points out their advantages and shortcomings. Then the classical algorithm of background modeling and updating in the background subtraction is introduced, including median filter, linear predictor, W4 method and eigen method. Finally based on the classical background modeling's shortcomings, this paper proposes a new algorithm for background modeling, called adaptive Gaussian mixture background modeling algorithm based on HSV color, which can fully use the effective information in the three color channels of HSV space. And when the body's shadow covered the around area or the light changes, we can automatically adjust background model learning rate through the HSV space's brightness channel V,so that the algorithm of the moving human object detection achieve closed-loop control and enhances the background modeling algorithm's robustness. Lastly we extract the moving human object through using this background modeling algorithm improved by Otsu algorithm and mathematical morphology.This paper did adepth research of tracking moving human object and abnormal human behavior recognition based on the moving human object detection. Firstly, we detect the moving human body's position and size by the human detection technology, and then describe both the horizontal and vertical direction and rotation of moving human objectives by introduction of the bandwidth matrix and angle of the moving human object. Lastly through dividing into rectangle blocks, we extract each block's kernel function to be weighted with gradient direction histogram,so that to achieve human object's multi-DOF tracking.Then this paper stores the moving human objective's characteristics such as height. width and the ratio of width to height are saved into the database, and then train using the SVM classifier, lastly use support vector machine classifier to identify whether the moving human's actity in the video stream is abnormal.