Pose Estimation Algorithm Derived From 3D Optical Flow

calculation method of 3D optical flow is proposed in this paper, which uses the aligned depth image and grayscale image to calculate the speed of pixel movement in 3D space. According to the obtained 3D optical flow, a direct normal pose estimation algorithm combining depth image information is proposed. The algorithm uses the gray value and the compensated depth value as the observation quantity to construct the least-squares problem, and solves the pose by means of graph optimization. The algorithm considers regions with large depth gradients and adds the compensated depth to the cost function. And according to the performance of 3D optical flow in the experiment, it is believed that the weight of the gray value should be higher than that of the depth value in the optimization problem. Compared with the direct method pose estimation algorithm, the algorithm proposed in this paper makes full use of the depth information. Experiments on the TUM dataset show that the improved pose estimation algorithm is more accurate than the original algorithm in scenes with better depth image quality, and the accuracy improvement is relatively more obvious in low-texture scenes.