3D target detection and tracking based on scene flow

This paper proposes a novel method for 3D multi moving object detection based on scene flow. Scene flow is solved by using the variational method. Datasets are captured by the stereo vision camera systems. Energy function consists of two parts - data term and smooth term. Data term includes the gray and gradient constancy assumption. Smooth term includes optical flow term and disparity term. We use the robust penalty function to reduce the impact of outliers and use multi-resolusion algorithm to solve large displacement problem. In order to solve the nonlinear problem that penalty function brings, we use fixed point iteration and first-order Taylor expansion. After the dence scene flow is obtained, we use Improved Fuzzy C-mean Clustering algorithm to segment the 3D motion vectors dynamically and obtain the moving target information. With the information of the camera, we can get the coordinate position of the moving object in the real world. Experiments show that our method can get better results of 3D moving target detection.