Skip to content
DSML.KZ
Publications

Article · Вестник КазАТК · April 28, 2023

NEURAL NETWORK DEVELOPMENT AND TRAINING FOR RECOGNIZING THE STATE OF A DRIVER

Olga Schwetz, Bauyrzhan Smakanov, Levente Kovac, Gjorgi George

Bauyrzhan Smakanov
Read on dx.doi.orgPDFDOI 10.52167/1609-1817-2023-125-2-186-195

Abstract

The topic of automated road safety support for all road users, the involvement of modern tools and advanced technologies to solve these problems is becoming more and more popular every year and does not lose its relevance. It was made a review of the currently existing works in this area, which showed the main advantages and disadvantages of the developments. It is noted the necessity to create a Kazakhstani automated system for monitoring the driver's condition, which will be available to the general public. The purpose of this study is to develop a new model based on a neural network for recognizing the state of a person driving a car to identify a dangerous state of the driver. The paper deals with the following issues: the relevance of the study, the formalization of the recognition problem and the choice of neural network architecture, which most successfully solves the problem. Separate freeze frames extracted from the video sequence of observing the driver are used for a mathematical description of the problem, a recognition function is built to select three classes corresponding to the driver's state, and feature vectors are selected to correctly determine the condition. For modeling, the architecture of a convolutional neural network was chosen, which allows working with images and video data streams. As a result of the study, a neural network was built and trained to recognize the dangerous state of the driver, taking into account all the factors influencing the process of identifying such a state. The created neural network included two convolutional layers describing the position of the head and the degree of eye openness, and also added two fully connected layers to take into account additional parameters, such as time of a day, duration of the trip, and whether the driver is yawning. Evaluation of modeling accuracy and comparison of the new model with existing ones shows the novelty and practical significance of this work.

Sleep and Work-Related FatigueHuman-Automation Interaction and SafetyAdvanced Neural Network Applicationsdriver state recognitiondangerous driving detectiondriver monitoring systems

More by Bauyrzhan Smakanov