Cross-view character recognition method based on shapes and postures under wearable equipment

Abstract

The invention belongs to the field of computer vision, relates to character recognition, and aims to better achieve character re-recognition with higher accuracy through human body posture information. In order to achieve the purpose, the invention relates to a cross-view character recognition method based on shapes and postures under wearable equipment. The method includes: giving a video frame image of a pedestrian to be detected in the first camera and a video obtained by the second camera; detecting a human body parameterized model and a two-position articulation point position corresponding to the frame image for all the video frames of the first two video frames, wherein the two-dimensional articulation point position is used for optimizing a human body parameterized model Smp1; obtaining a final human body parameterization model through three-dimensional joint point reprojection optimization operation and discrete cosine transform (DCT) time domain optimization operation of thehuman body parameterization model; and comparing the final human body parameterized model of the target to be detected with the human body parameterized model in each video frame in the second camera, and finding out the target to be detected. The method is mainly applied to automatic character recognition occasions.

Description

technical field [0001] The invention belongs to the field of computer vision, human body pose estimation, and model optimization, and relates to a cross-viewpoint character recognition method based on 2D / 3D human body joint points, a parametric model of a human body and optimization. Background technique [0002] As a sub-problem of image retrieval, the research of cross-view person recognition (CVPI) is very important and has a wide range of application scenarios, such as: video surveillance in outdoor dense areas, intelligent human-computer interaction and military investigation. Early research can be traced back to the problem of cross-camera multi-target tracking. Traditional surveillance videos are mostly obtained through fixed-position cameras. They can only cover a limited area from a pre-fixed perspective. Due to the fixedness of the camera, it will cause occlusion and Problems such as the disappearance of the target pedestrian’s field of view in the camera, so early...

AI Technical Summary

Problems solved by technology

Due to the inaccuracy of the pose estimated by this method, and the detection set is input through consecutive frames of video, this method ignores the characteristics of pose consistency in consecutive frames of video, which will affect the accuracy of CVPI to a certain extent.

Images