3d human pose estimation based on multi view information fusion

3D human pose estimation refers to recovering the pose of the human body in the three-dimensional space from the image, which is estimating the coordinates of the key points of the human body in the three-dimensional coordinate system. The human skeleton model composed of these keypoints describes the human pose in the image.The traditional pose estimation method uses motion capture equipment, which requires athletes to wear auxiliary equipment and affects the normal training of athletes. With the development of deep learning, convolutional neural network has shown strong representational capabilities in the process of image feature extraction, and is an effective method to achieve human pose estimation. However, in the actual complex and changeable three-dimensional space, the environment is noisy and the limbs are occluded. The monocular capture of human posture has the problems of low accuracy and instability, and cannot provide sufficient semantic information.To solve this poroblem, this paper designs a multi-view information fusion algorithm. First, a two-dimensional attitude feature extraction module is designed. By connecting and interacting heat maps of different resolving power in parallel, the heat map always maintains a higher resolving power for higher precision. It extracts the two-dimensional human pose from a single perspective, and then uses the multi-view information fusion technology proposed in this paper to complete and correct the pose information from different perspectives to obtain a more accurate two-dimensional pose. Then, through the camera pose transformation, the three-dimensional human pose estimation is realized, and the experimental verification is carried out on the multi-eye machine vision pan-tilt zoom tracking shooting system.