Human body key point detection method and system based on spatial transformation information transmission

Abstract

The invention relates to the field of computer vision, in particular to a human body key point detection method and system based on spatial transformation information transmission, thereby reducing generation of null prediction and false positive prediction. The detection method comprises the steps: inputting a to-be-detected human body image into an adversarial network, extracting key point features and limb stem features, and then generating a key point first heat map and a limb stem heat map; downsampling the to-be-detected human body image to obtain a downsampled human body image; and according to the down-sampled human body image, the key point first heat map, the limb heat map and the key point features, generating a key point heat map after information transfer correction based on aspatial transformation network. According to the method, prediction priori of the human body key points is supplemented through spatial relation transformation and transmission of the adjacent key points, noises brought to key point prediction by shielding and human body posture diversity are restrained, so that generation of null prediction and false positive prediction is effectively reduced.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a human body key point detection method and system based on spatial transformation information transmission. Background technique [0002] Human body key point detection is one of the hot research topics in computer vision, and it is also an important direction for the application of vision-based pattern recognition. It needs to detect key points such as ankles and wrists of people in the picture. It is the basis of many high-level visual tasks, such as behavior recognition, clothing analysis, pedestrian re-identification, human-computer interaction and a series of tasks. However, in actual application scenarios, the skeleton structure of the human body is not completely rigid, and various and complex human postures make it very difficult to estimate key points; in addition to being affected by complex and changeable human postures, self-occlusion often occurs and object occlusion...

AI Technical Summary

Problems solved by technology

[0004] However, there are still many shortcomings in the human body key point detection method based on deep learning.

Due to the occurrence of problems such as occlusion and attitude changes, network prediction will produce great ambiguity, resulting in the network being unable to detect the exact position of key points, and many misjudgments occur.

Images