Shadow elimination method in personnel image detection in electric power practical training field

Abstract

The invention provides a shadow elimination method in personnel image detection in an electric power practical training field. The method comprises the following steps: sample data is obtained; the discriminator judges whether the shadow image is generated by the generator or a real shadow image, a first error between the shadow image and the real shadow image is fed back to the generator, and the generator reduces the first error; the discriminator judges whether the shadow-free image is from the generator or a real shadow-free image, a second error between the shadow-free image and the real shadow-free image is fed back to the generator, and the generator reduces the second error; the generator and the discriminator carry out repeated adversarial training to obtain an optimal shadow removal network model; and shadow removal is performed by using the shadow removal network model, thus inputting a shadow-containing personnel image in the electric power practical training field, and outputting a shadow-free personnel image in the electric power practical training field. Through massive training of a shadow data set, the neural network is enabled to continuously perform self-gaming in confrontation, so that style migration of a shadow image is better realized.

Description

technical field [0001] The invention relates to a method for removing shadows from images, in particular to a method for removing shadows in image detection of persons in an electric power training field. Background technique [0002] Video analysis-based moving image understanding has always been a hot spot in the computer field. As early as 2005 in the field of safety production, someone proposed a helmet wearing state detection algorithm in a paper: helmet detection based on the feature of the gradient direction histogram, or the method of skin color detection first detects the face area, and then obtains the face area. The helmet is detected by feature extraction after the partial image. In addition, some people have proposed compliance detection of personal protective equipment in construction sites, safety inspection of live video based on computer vision methods, or helmet wearing detection through deep learning YOLO-v3 network. [0003] In the field of electric pow...

AI Technical Summary

Problems solved by technology

[0004] However, due to the complexity of the video scene for power grid monitoring and the frequent changes in the posture of the operator, the problem of recognition accuracy of the operator will be greatly reduced

In order to improve the safety supervision level

Images