Research On Detection And Gesture Recognition Algorithm For Moving Target Based On Vision

With the worldwide aging society is becoming more and more serious and theconstruction of ‘harmonious society’, improve the quality of life of the elderly is theurgent need of monitoring of elderly people living alone, the more and more important.Vision is one of the human perception of its complicated environment around the mostdirect and effective means, but in real life, a lot of meaningful information of vision areincluded in the movement, but also more sensitive to the human eye to moving objectsand goals, can quickly find the moving target, and the target moving trajectoryprediction and description. In this paper, through the analysis of the video content usingcomputer vision technology, detection, tracking and abnormal posture recognition ofhuman motion algorithm, through the test the effect of simulation experiment, finallyrealize the intelligent monitoring of the elderly living alone. Study specific algorithminvolves moving target detection, tracking and human abnormal posture detecting partthree.First of all, in moving target detection, the moving target detection method basedon inter frame difference method, through the gray scale, the adjacent image difference,two values, morphological filtering, connectivity detection steps, finally completemoving object region extraction. The inter frame difference method is simple, easy toimplement for real-time moving object detection, also need not consider the backgroundupdating, insensitive to illumination changes, shadows are also relatively small.Secondly, in the respect of target tracking, first introduced the tracking methodcurrently used, application field of each algorithm, the moving target tracking methodbased on Kalman filter. Kalman filter is a tracking algorithm based on Bayesian estimation theory, using Kalman filtering system estimation model, the optimalestimation of moving target, and the simulation results on the performance of thealgorithm is verified, and receives good effect.Finally, in the abnormal posture recognition, first introduced the current abnormalbehavior analysis; in moving target identity, we use the target mass outer rectangle logo,the centroid coordinates and height to width ratio feature extraction; in the human falldetection method, the centroid coordinates and the outer rectangle high aspect ratio isjudged; at the end of the indoor environment of human body fall was tested byMATLAB simulation. Good results.