Real-time 3D human body posture estimation method in transformer substation scene

Abstract

The invention relates to a real-time 3D human body posture estimation method in a transformer substation scene, which realizes human body 3D joint point extraction in a transformer substation operation scene based on top-down and bottom-up method fusion and a 3D posture estimation network of a multi-layer Transform network architecture, proposes a strategy based on top-down and bottom-up method fusion, and realizes real-time 3D human body posture estimation in a transformer substation operation scene. The 2D joint point result similarity generated by the top-down method and the bottom-up method is calculated and compared with a manually set threshold value, fusion of the top-down method and the bottom-up method is achieved, and the robustness of 3D human body posture estimation is improved. According to the method, the Transform is used for replacing a convolutional neural network, the time sequence information and the space information of the input image are obtained by using the self-supervision capability of the Transform, the estimation capability of the occlusion target is improved, and the accuracy of human body 3D joint point estimation is further improved.

Description

technical field [0001] The present invention relates to the technical field of methods for estimating real-time 3D human body postures, in particular to a method for estimating real-time 3D human body postures in a substation scene. Background technique [0002] The working scene of the substation is complex, and there are a large number of unsafe factors, such as illegal operation and improper use of tools, etc., which can easily cause accidents. At present, the substation operation site generally lacks effective human motion recognition technology, and it is difficult to detect and stop violations in time only through human eye video monitoring. The 3D joint point information of the human body is an important underlying information for human action recognition. If the artificial intelligence technology can be used to accurately estimate the human body posture of the operator and determine the position of the 3D joint point, it will play a role in the discovery and early wa...

AI Technical Summary

Problems solved by technology

However, no matter whether it is LSTM or graph convolutional network, the calculation consumption is large when the time span is large and the network is deep. Efficient estimation is the main problem faced by current pose estimation methods at the application level

Images