Eye and mouth state recognition method based on convolutional neural network

Abstract

The invention relates to an eye and mouth state recognition method based on a convolutional neural network. SR-Net (State recognition nets) designed through the method learn a large amount of different eye and mouth state samples. Face state recognition can be regarded as eye state recognition and mouth state recognition. Eye and mouth states can be recognized and classified more accurately. Due to the convolutional neural network, extraction of artificial features is avoided, and recognition on eye and mouth states has high robustness. Due to the method of the invention, the recognition rate in a sunglass wearing condition is enhanced, the average accuracy for eye state recognition is improved to more than 98.41%, the average recognition rate for eye states without glasses is 98.92%, and the average recognition rate for the mouth states is 99.33%.

Description

technical field [0001] The invention relates to an eye and mouth state recognition method based on a convolutional neural network. The method can adapt to illumination changes and glasses blocking situations, belongs to the technical field of image recognition, and can be applied to determine the driver's fatigue state. Background technique [0002] Eye and mouth state recognition can be considered equivalent to the recognition of human face and face state. It is an important content in the field of image recognition and has a direct impact on information security, automatic driving and other technologies. According to the report of the National Center for Statistics and Analysis of the United States, fatigue driving is one of the important causes of traffic accidents. Therefore, the research on driver fatigue detection technology has important significance for preventing traffic accidents. In recent years, with the improvement of computer hardware level, the fatigue detecti...

AI Technical Summary

Problems solved by technology

Using multiple template matching, the detection efficiency is low and the real-time performance is poor

The second type of method uses the gray projection curve of the iris area of ​​the eye to judge the state of the eye, which has higher requirements on lighting and is less applicable to the scene.

The third type of method uses eye opening and closing detection based on the combination of LBP features and SVM classifiers, which has certain limitations for drivers wearing sunglasses and posture changes, and has poor robustness.

The fourth type of method uses eye state recognition based on multi-feature fusion. This method requires multiple classifiers for decision fusion, and the real-time performance is poor.

[0005] The convolutional neural network has better expressive power for features and avoids the manual feature selection process. Before the convolutional neural network was proposed, due to the lack of similar technical means, the field of image recognition has been limited to "extract features first, then pattern recognition". framework, the accuracy of recognition cannot break through the bottleneck of the technical framework, and the progress is slow

Images