NEURAL NETWORK SYSTEM FOR DETERMINING HUMAN DROPPINESS BY FACIAL EXPRESSION
Authors:
1.
Kazan National Research Technical University named after A.N. Tupolev
2.
KNITU-KAI
3.
Kazanskiy nacional'nyy issledovatel'skiy tehnicheskiy universitet imeni A.N. Tupoleva
4.
Kazan National Research Technical University named after A.N. Tupolev
Type:
Article
Pages:
from 158
to 164
Status:
In work
Language:
Russian
Keywords:
NEURAL NETWORK SYSTEM FOR DETERMINING HUMAN DROPPINESS BY FACIAL EXPRESSION
Abstract (English):
The article is devoted to the development of a neural network system for determining human drowsiness by facial expression. There are various drowsiness detection systems that track a person's driving style, driver's physiological indicators, and recognize facial expressions. However, these systems are often unavailable to users due to their high cost, which makes it necessary to develop our own system. The analysis of the subject area showed that human drowsiness assessment can be based on the use of EAR (Eye Aspect Ratio) and MAR (Mouth Aspect Ratio) indicators. The first indicator determines the degree of eye openness, and the second determines the degree of mouth openness. These characteristics can be calculated by using a special face mask on an image with key points to detect a person's mouth and eyes. The algorithm for recognizing drowsiness by facial characteristics includes the following steps: capturing a face image, localizing the eyes and mouth, calculating EAR and MAR, determining threshold values, monitoring and recognizing drowsiness. Despite many applications, the use of this technology to determine human drowsiness requires additional research. Of particular relevance in this area is the construction and study of convolutional neural network models and systems. To build such a system, it is necessary to prepare data for analysis, build a neural network convolutional model, implement a graphical interface for the neural network system and check the adequacy of its operation. In this study, we used the Driver Drowsiness Dataset, Drowsiness Prediction Dataset and UTA Real-Life Drowsiness Dataset obtained from the publicly available Kaggle source. The YOLO neural network version v5 was used to analyze the data. When building a neural network model on the Google Colab platform, the Python programming language was used. The model training process lasted for 30 epochs. As a result of training, the accuracy of determining the vigorous class was 97.9%, the accuracy of determining the sleepy class was 95.7%. The average accuracy of the model was 96.8%, which is a high result. Based on the constructed model, a neural network system was developed in the Visual Studio environment. The Python programming language was used for its development. To evaluate the effectiveness of the developed system, it was validated using the following classification quality metrics: accuracy, recall, F1-measure and precision. The results of the validation showed that the system classifies most cases of human drowsiness and alertness quite accurately. This indicates its effectiveness and the possibility of solving practical problems.
The article is devoted to the development of a neural network system for determining human drowsiness by facial expression. There are various drowsiness detection systems that track a person's driving style, driver's physiological indicators, and recognize facial expressions. However, these systems are often unavailable to users due to their high cost, which makes it necessary to develop our own system. The analysis of the subject area showed that human drowsiness assessment can be based on the use of EAR (Eye Aspect Ratio) and MAR (Mouth Aspect Ratio) indicators. The first indicator determines the degree of eye openness, and the second determines the degree of mouth openness. These characteristics can be calculated by using a special face mask on an image with key points to detect a person's mouth and eyes. The algorithm for recognizing drowsiness by facial characteristics includes the following steps: capturing a face image, localizing the eyes and mouth, calculating EAR and MAR, determining threshold values, monitoring and recognizing drowsiness. Despite many applications, the use of this technology to determine human drowsiness requires additional research. Of particular relevance in this area is the construction and study of convolutional neural network models and systems. To build such a system, it is necessary to prepare data for analysis, build a neural network convolutional model, implement a graphical interface for the neural network system and check the adequacy of its operation. In this study, we used the Driver Drowsiness Dataset, Drowsiness Prediction Dataset and UTA Real-Life Drowsiness Dataset obtained from the publicly available Kaggle source. The YOLO neural network version v5 was used to analyze the data. When building a neural network model on the Google Colab platform, the Python programming language was used. The model training process lasted for 30 epochs. As a result of training, the accuracy of determining the vigorous class was 97.9%, the accuracy of determining the sleepy class was 95.7%. The average accuracy of the model was 96.8%, which is a high result. Based on the constructed model, a neural network system was developed in the Visual Studio environment. The Python programming language was used for its development. To evaluate the effectiveness of the developed system, it was validated using the following classification quality metrics: accuracy, recall, F1-measure and precision. The results of the validation showed that the system classifies most cases of human drowsiness and alertness quite accurately. This indicates its effectiveness and the possibility of solving practical problems.
Keywords:
NEURAL NETWORK SYSTEM FOR DETERMINING HUMAN DROPPINESS BY FACIAL EXPRESSION
NEURAL NETWORK SYSTEM FOR DETERMINING HUMAN DROPPINESS BY FACIAL EXPRESSION
