Detection And Tracking Of Human Objects Based On Vision

With the advancement of informatization and intelligence,the demand for computer vision analysis in many fields of daily life and work will also increase day by day.However,human recognition and tracking technology are important research directions in the field of computer vision.There are wide applications in intelligent transportation systems,robots,interactive systems,intelligent video surveillance systems,autonomous driving systems.In this paper,we mainly study human target recognition and moving human target tracking.Color image and depth image acquisition are collected by Kinect sensor of Microsoft somatosensory device.The main work of this paper includes the following two parts:Human recognition based on Kinect sensor color image and depth image is realized by improving the method of human target recognition in color image and combining with depth image information.The HOG(gradient direction histogram)feature is extracted from the color image,the HOD(depth direction histogram)feature is extracted from the depth image,and the integral feature map and the three-line interpolation method are used to simplify the acceleration feature extraction process.Afterwards,the two kinds of features are respectively introduced into the SVM classifier for model training.Finally,the two classifiers fuse for effective human detection.Because the depth image is less affected by the light factors,it can be accurately obtained even in the dark environment.Also,the spatial distance information reflected by the depth information helps to solve the occlusion problem.Experimental results show that the proposed method enhances the robustness of vision-based human target recognition to illumination and the detection accuracy under occlusion.In moving human target tracking,we focus on moving target tracking method based on Kalman filter prediction and tracking method based on Camshift algorithm.Kalman filter-based tracking is to predict the target's motion information and based on Camshift algorithm's tracking is to use the target's own color characteristics for color matching tracking.This article combines the two to achieve the tracking of the human body's moving target in the video image.In the event of occlusion of the target or loss of tracking,Kalman filter is used to search for the target within a certain area and improve the probability of reaching the human target of exercise again.After searching the target again,Kalman filter is corrected by using the tracking result of Camshift algorithm.This method combines the color information of the target with the motion information to ensure the accuracy of the tracking and can largely solve the problem of obscuring the target.