Human skeleton key point detection method and device

Abstract

The invention discloses a human skeleton key point detection method and device. The method comprises the following steps that step one, a monocular depth map is acquired; and step two, the monocular depth map is inputted to a trained full convolutional deep neural network so as to detect the human skeleton key point position, wherein the network feed-forward low abstract early feature map and thehigh abstract later feature map are integrated by the full convolutional deep neural network, and the small feature map is enlarged to other feature maps by using up-sampling so that global information including the small feature map and feature information including the large feature map are inputted in the next convolution computation. The feature information can be effectively utilized with combination of the early and later feature information and the global and local information so that the more abundant feature information can be outputted and the skeleton key point locating effect can be enhanced.

Description

technical field [0001] The invention relates to a method and a device for detecting key points of human bones, in particular to a full convolutional neural network using front and later feature map information, local and global information to detect the positions of key points of human bones in a two-dimensional depth map. Bone key point detection method and device. Background technique [0002] One of the important tasks in the field of computer vision research is the detection of human skeleton key points, specifically to enable the computer to perceive the position of each skeleton key point of the human body, for further action recognition, abnormal action detection, intelligent monitoring, automatic driving, etc. A practical scenario provides the basis. [0003] The goal of the human skeleton key point detection task is to take a picture as input, and output the horizontal and vertical coordinates of each skeleton key point of the human body in the picture. There are ...

AI Technical Summary

Problems solved by technology

[0004] However, due to the lack of records of human body details in the depth map, there will be some depth noise in the depth map, and difficulties such as human body self-occlusion and external obstacle occlusion still exist, so it is required to perform two-dimensional human skeleton key points based on depth map stably and quickly Detection is still a very challenging topic

Images