A 2D Pose Estimation Method for Video Sequences Based on Reinforcement Learning
A pose estimation and reinforcement learning technology, applied in the field of video two-dimensional human pose estimation, can solve the problem of inaccurate pose estimation and achieve the effect of improving accuracy
Active Publication Date: 2022-06-07
WUHAN UNIV
View PDF10 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0007] The present invention proposes a two-dimensional attitude estimation method for video sequences based on reinforcement learning, which is used to solve or at least partially solve the technical problem of inaccurate attitude estimation in the methods of the prior art
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0052] The embodiment of the present invention provides a two-dimensional pose estimation method for a video sequence based on reinforcement learning, which is used to improve the technical problem that pose estimation is not accurate enough in the prior art method.
[0053] The inventor of the present application found through a large amount of research and practice that the existing methods mainly have the following two technical problems:
[0054] First, the timing information mining in the video is not sufficient. The existing methods only focus on the local information of human motion in the video, but ignore the modeling of the overall motion process. When the self-occlusion and motion blur of the human body appear in the video, if the global information is not considered, the performance of the existing method will be reduced. will be significantly reduced.
[0055] Second, it is not effective enough to utilize a small amount of pose label information in the video. Th...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The present invention provides a two-dimensional pose estimation method for a video sequence based on reinforcement learning. First, the overall state of the reinforcement learning agent is constructed, and then an action is output. A marked frame and an unmarked frame are selected from the current video sequence, and the two frames pass through The attitude estimator obtains the attitude, and is sent to the action converter to complete the action conversion and update the attitude estimator and action converter. Finally, the reward of the agent is calculated according to the improvement of the attitude estimator, and the optimized agent is updated. The present invention utilizes the idea of reinforcement learning and active learning to select the most informative video frame pair by fully mining the timing information in the video, so that it can well resist the problems of human body self-occlusion, motion blur, etc., and can improve the estimation of video pose accuracy and robustness of the method. The present invention is different from common supervised learning schemes, and only uses a small amount of video tagged frames to complete the learning of the attitude estimator, which significantly reduces the amount of manual tagging, thus having higher practicability.
Description
technical field [0001] The invention relates to the technical field of video two-dimensional human body pose estimation, in particular to a two-dimensional pose estimation method for video sequences based on reinforcement learning. Background technique [0002] Video human pose estimation has always attracted the attention of researchers in the computer vision field. It is an important research direction in the field of computer vision. Its core research content is to detect human bodies (including motion segmentation and target classification) from single or multiple video sequences. Tracking and recognition and understanding of human motion (including pose estimation and action recognition and their description). Among them, human pose estimation in monocular video is one of the most complex research branches in video human pose estimation, in order to analyze human image features from monocular video, and then estimate the two-dimensional human pose parameters. Because i...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G06V40/10G06V20/40G06V10/774G06K9/62G06N20/00
CPCG06N20/00G06V40/23G06V20/42G06F18/2155
Inventor 陈军马宪政刘涛榕常路
Owner WUHAN UNIV