A Human Fatigue Detection Method Based on Improved Cascaded Convolutional Neural Network

Abstract

The present invention claims to protect a human body fatigue detection method based on an improved cascaded convolutional neural network. A cascaded neural network structure is designed to detect human eyes and human eye feature points, and the first-level network uses gray scale integral projection rough positioning and multi-task convolutional neural network (G‑RCNN) network to realize human eye detection and positioning , the second-level network (PCNN) divides the human eye picture and uses a parallel sub-convolution system to perform feature point regression prediction; use the human eye feature points to calculate the human eye opening and closing degree to identify the current human eye state; S4, judge the human body according to the PERCLOS criterion Fatigue state; the present invention can obtain a higher recognition rate, and can have stronger robustness to illumination and random noise.

Description

Technical field [0001] The present invention belongs to the field of image processing and mode identification, particularly a human fatigue detection method based on an improved cascaded neural network. Background technique [0002] Fatigue refers to the state of labor efficiency due to excessive long-term or over-tension due to excessive long or over-tension, and mental fatigue is the origin of multiple conditions. Fatigue not only harms the physical and mental health of people, but also brings major safety hazards to social production and life, especially in high-risk operations such as electricity industry, building high-altitude operations, vehicles driving, aerospace, large complex industries, and production control personnel. Fatigue caused attention to dispersion, slow response or decline in body coordination, which may lead to extremely serious production accidents. In terms of car driving, with the increase in China's total, traffic accidents are also more frequent, and ...

AI Technical Summary

Problems solved by technology

Deng et al. used the skin color model combined with the layout of the three courts and five eyes of the face to locate the human eye, and used the size of the integral projection area of ​​the human eye to identify the state of the human eye. Although this method is simple in algorithm, the accuracy of positioning is less affected by the environment. Large, and because the proportion of the human eye area in the image is very small, the recognition accuracy of the human eye state using the integral projection is low

Li Xiang et al. use the moment feature of the image to calculate the similarity between the Zernike moment feature vector of the human eye template and the face area to be recognized, and select the area with the largest similarity as the human eye area. This method is similar to the template matching method, although it can reduce The impact of the small environment on the detection results, but the calculation is relatively large, and the results are affected by the selected human eye template

Images