Depth Image Extraction And Analysis Based On Binocular Video Capture

China has stepped into an aging society.With the developing of economy and more people go out to work,the number of elderly people living alone is increasing.The importance of geriatric guardianship system is becoming increasingly prominent.Traditional guardianship systems usually rely on guardians,which are time-consuming and laborious.With the development of technology,new sensors and intelligent devices will become the focus point of research.For example,some contact sensors can extract biological information to monitor the state of the elderly;and some monitoring cameras can monitor 360 degrees.But the contact sensors rely on people's willingness,and 360-degree cameras can not break through the limitations of image information.In order to solve these shortcomings,a binocular camera is introduced into the monitoring system to realize an embedded platform for extracting three-dimensional image information.In the process of studying binocular stereo camera,several applications of minimum spanning tree in stereo matching are summarized and studied.Minimum Spanning Tree(MST)can generate similar relationships between pixels,which can help to repair the depth map that depth information cannot be computed.By testing the data set provided by Middlebury website and using the results of BM algorithm and SGBM algorithm as the images to be repaired,the mismatch rate can be reduced by more than 10%.In the process of stereo matching cost aggregation,the minimum spanning tree can construct global connection based on all the pixels to complete the cost aggregation.Compared with BM and SGBM,the mismatch rate and root mean square error of MST cost aggregation are also reduced.Aiming at the application of binocular camera in monitoring field,the transplantation of ARM embedded platform based on binocular algorithm is completed.The embedded platform connects the binocular camera through the USB interface and displays the result through the display screen.The software part uses the Linux embedded operating system to development,and supports the algorithm program by transplanting OpenCV library and Qt display framework.Finally,the human action image is extracted from the indoor environment to test the platform.