Study On Three-dimensional Trajectory Reconstruction Of Moving Object Based On Binocular Vision

No matter video monitoring or observation of microorganisms in biological or medical conditions, missile launch in military field, navigation system of mobile robot and other fields, always involving the tracking and location of moving object, and to further to understand its behavior, also that requires us to reconstruct3d trajectory of moving object. The main problems of study on three-dimensional trajectory reconstruction of moving object based on binocular vision include moving object detection, target tracking and3d reconstruction. This thesis focused on building up binocular stereo vision system, target detection and tracking for Image sequence which have been taken from binocular vision. After known the2d trajectory of moving object, the three-dimensional trajectory will be required through projection matrix. The main achievements of this thesis include:(1) Binocular stereo vision system is established, analyzed common coordinate system in the binocular vision coordinate system and transformation equation between each coordinate system. Then the camera model is analyzed and discussed, and obtained the internal parameters and the external parameters of the camera model using Zhang zhengyou calibration method which preparing for the three-dimensional trajectory reconstruction.(2) The thesis described the current methods for moving object detection. On the basis of frame difference, using edge detection, contour extraction and segmentation method to detect the moving object. Secondly, this paper expounds the current commonly used several tracking algorithms, according to the need of target tracking, Camshift algorithm is selected as one of the main tracking algorithm. The tracking algorithm is analyzed, tracking results are given when there are some influencing factors which include background interference, block appeared, and the speed of moving object is too fast and so on in the processing of tracking. Some plans are given to solve these problems.(3) As camshift algorithm is not very well used in target tracking, particle filter algorithm is introduced. Firstly, this paper introduced the basic theoretical knowledge of particle filter algorithm. Secondly, the particle filter is applied to track moving targets, and according to these problems of the algorithm, the thesis proposed the adaptive state prediction model of particle filter tracking algorithm, it is able to solve inaccuracy of tracking results when the target obscured and sports autonomy problem. Finally, the Camshift algorithm is embedded into the improved particle filter algorithm, the method effectively overcome some problems of large amount of calculation and poor real-time performance. Experiments verify that the method has a good tracking effect.(4) On the basis of the previous chapters, the two-dimensional trajectory have been obtained and known the internal and external parameters of two cameras, three-dimensional trajectory of moving object has been obtained through least square method. In order to analysis the error of trajectory, considering placed the checkerboard in different locations in the world coordinates, compared these true location through measuring and the location which be from system, the average error is small in X and Y direction, with the increase of depth distance, the error is growing and nonlinear on the Z axis.Experimental results show that the error is within acceptable range and the system can meet practical application that do not need strict requirements on accuracy.