Research On 3D Object Tracking Algorithm Based On Parameter Adaptive Of Particle Filter

In recent years,with the decline in computer storage costs and more and more broad application prospects in the market,moving object tracking has become a hot topic in the field of computer vision.Low-cost point cloud(RGBD)acquisition equipment also appears in the market.Compared with the two-dimensional image,the point cloud image contains depth information,will be not affected by light.It often used for moving object detection and tracking research.The research of 3D object tracking is becoming more and more significant.In order to realize real-time multi object tracking in 3D environment,thesis firstly design 3D single-point tracking system,which uses KLD particle filter algorithm to track single object in single object tracking.The down-sampling filter parameters is closely related to the tracking target size.The maximum particle number,error bound and bin-size are closely related to the number of point clouds.Inaccurate parameter settings can result in poor tracking or even failure.In this system,thesis studied the parameters of the tracking process selection rules,the correct order of the parameters and parameters setting rules through a large number of experiments.According to the rules,parameters can be adaptively set reasonably based on the object size and point cloud number.Each tracking object can achieve a better tracking effect.The main work of thesis includes the following parts.Firstly,in the moving object tracking system,the current scene needs to be segmented to determine the initial position of the object.A regional growth segmentation algorithm based on color and normal is proposed.The kd-tree in PCL library is used to search the nearest neighbor,which reduces the time spent by the algorithm and improves the segmentation efficiency.Secondly,after the object segmentation,the object color and distance are extracted,and the corresponding similarity measure function is established to update the weight ofthe particle in the KLD particle filter algorithm,and the KLD particle filter object tracking in PCL environment is realized.Thirdly,under the single-objective tracking system,the relationship between the parameters and the tracking object is analyzed by a large number of experiments.The correct order and the rule of the parameters are obtained,and the multi-moving object tracking is achieved.