Human body key point detection and self-learning method and device

Abstract

The invention provides a human body key point detection and self-learning method and device. The method comprises the steps that image information of multiple continuous frames is extracted through cameras or videos, multiple key points of a human body are detected for each image, wherein the key points have coordinate information and confidence coefficient information, and for the key points withthe confidence coefficient smaller than 0.6, an adjacent edge curve fitting algorithm is used for completing coordinate prediction and confidence coefficient calculation of the key points with the confidence coefficient smaller than 0.6. According to the method, the coordinates of the shielded key points of which the confidence coefficients are smaller than 0.6 are predicted, so that the accuracyof human body key point detection is improved, and human body key point information can be returned more accurately.

Description

technical field

[0001] The invention relates to the technical field of image processing, in particular to a human body key point detection and self-learning method and device. Background technique

[0002] In recent years, with the development of computer vision technology, the field of human body key point prediction has made great progress in image processing. A keypoint is essentially a feature. It is an abstract description of a fixed area or spatial physical relationship, and describes the combination or contextual relationship within a certain neighborhood. It is not just a point of information, or represents a location, but also represents the combination of up and down and surrounding neighbors. The existing human body key point detection technology often leads to wrong prediction of 3D human body pose due to the wrong depth prediction of human body key points.

[0003] The Chinese patent application publication number is CN108830139A, and the title of the inventi...

Examples

Embodiment 1

[0043] The present invention extracts the image information of a plurality of continuous frames through a camera or video, and for each image, uses the human body posture detection algorithm Openpose to detect a plurality of key points of the human body, these key points have coordinate information and confidence degree information, for the confidence degree Low key points, mostly caused by occlusion. These key points with low confidence and zero confidence make it difficult for subsequent action recognition or behavior analysis. Therefore, it is necessary to improve the prediction of the coordinates of the partially occluded key points of the human body, increase their confidence, and return the key points of the human body as accurately as possible. information.

[0044] For key points with higher confidence (value greater than 0.9), keep the original value, and for key points with low confidence (value less than 0.6), design a low-confidence key point prediction algorithm, ...

Embodiment 2

[0072] This embodiment provides a human key point detection and self-learning device, such as figure 2 As shown, including acquisition module, preprocessing module, detection module, calculation module and learning module;

[0073] Acquisition module: used to extract each frame image of the stream data of the camera or video and send the extracted image to the preprocessing module;

[0074] Preprocessing module: preprocessing each frame of the extracted image, and sending the preprocessed image of each frame to the detection module.

[0075] With the help of a cache queue, using images of multiple frames before and after (here 5 frames are taken, and 2 frames before and after the current frame), according to the principle that the weight of the frame image farther away from the current frame is lower, complete the image based on the corresponding pixel. The weighted average method obtains the processed image of the current frame.

technical field

[0001] The invention relates to the technical field of image processing, in particular to a human body key point detection and self-learning method and device. Background technique

[0002] In recent years, with the development of computer vision technology, the field of human body key point prediction has made great progress in image processing. A keypoint is essentially a feature. It is an abstract description of a fixed area or spatial physical relationship, and describes the combination or contextual relationship within a certain neighborhood. It is not just a point of information, or represents a location, but also represents the combination of up and down and surrounding neighbors. The existing human body key point detection technology often leads to wrong prediction of 3D human body pose due to the wrong depth prediction of human body key points.

[0003] The Chinese patent application publication number is CN108830139A, and the title of the inventi...

Examples

Embodiment 1

[0043] The present invention extracts the image information of a plurality of continuous frames through a camera or video, and for each image, uses the human body posture detection algorithm Openpose to detect a plurality of key points of the human body, these key points have coordinate information and confidence degree information, for the confidence degree Low key points, mostly caused by occlusion. These key points with low confidence and zero confidence make it difficult for subsequent action recognition or behavior analysis. Therefore, it is necessary to improve the prediction of the coordinates of the partially occluded key points of the human body, increase their confidence, and return the key points of the human body as accurately as possible. information.

[0044] For key points with higher confidence (value greater than 0.9), keep the original value, and for key points with low confidence (value less than 0.6), design a low-confidence key point prediction algorithm, ...

Embodiment 2

[0072] This embodiment provides a human key point detection and self-learning device, such as figure 2 As shown, including acquisition module, preprocessing module, detection module, calculation module and learning module;

[0073] Acquisition module: used to extract each frame image of the stream data of the camera or video and send the extracted image to the preprocessing module;

[0074] Preprocessing module: preprocessing each frame of the extracted image, and sending the preprocessed image of each frame to the detection module.

[0075] With the help of a cache queue, using images of multiple frames before and after (here 5 frames are taken, and 2 frames before and after the current frame), according to the principle that the weight of the frame image farther away from the current frame is lower, complete the image based on the corresponding pixel. The weighted average method obtains the processed image of the current frame.

Images