The Research On Kinect-based Motion Detection System

Motion detection system is one of the most basic and most important aspects in the field of computer vision research.The author has developed a main research project of The Ministry of Public Security which has proposed high detection rate,high real-time and accurate contour marking and other requirements for the performance indicators of indoor object motion detection system.In the process of system development and application,it is found that the traditional motion detection system based on monocular camera still has the following deficiencies in practical applications:1)Motion detection system based on single color image is subject to large background interference,and there are inevitable problems such as false detection,leak detection and under-detection and other problems.2)The motion detection system contains many digital image processing steps,and the program nesting process may result in slow operation speed and display jam and other problems.3)Single color information cannot meet the current diverse software function development needs.In order to solve the above problems,this dissertation chooses to replace the traditional camera with 3D camera Kinect and develops a Kinect-based motion detection system.The system is designed for indoor environment,focusing on motion detection of all objects in the scene,displaying the specific starting position and precise contour of the moving target in real time,and using the de pth information to establish three-dimensional space coordinates,so as to does spatial motion analysis and 3D reconstruction of scene of the target.Firstly,from the original image requirements,the color image and the depth image collected by Kinect are accurately matched,and the fast Gaussian transform is introduced for the two indexes of edge retention and operation speed which optimize and improve the LMS algorithm and the LBF algorithm,so as to achieve a fast denoising method for deep image edge retention.Secondly,based on the GMM algorithm,the motion detection of the fused image sequence collected by Kinect is carried out,and the motion detection algorithm is speeded up by rationally optimizing the program structure,reducing the number of scans and simplifying the mathematical model.Thirdly,based on the GrabCut algorithm,the target foreground is segmented,and the corresponding mathematical model is established by using the spatial transformation relationship of the coordinate system to anal yze the spatial motion of the target.The three-dimensional model of the scene is established in the form of a point cloud map to represent the target's displacement situation in three-dimensional space.Finally,based on VS 2015 and Qt,this dissertation implements and tests the “Kinect-based motion detection system”,which combines the use of multiple complex scenarios to evaluate and summarize the software performance.