Highway Drivers Drowsiness Detection System Model with R-Pi and CNN technique

Abstract

In order to drive the state of art technology for driver safety, the demand for Artificial Intelligence and Machine learning precision approaches are popularly used in the modern vehicles. This paper presents a driver drowsiness detection framework which can be placed in the vehicle as a standalone design unit. In this organised deep learning CNN frame work, a comprehensive analytics strategy is used to detect drowsiness in drivers. There are two distinct phases used in this experimental framework namely image data acquisition followed by identification of region of interest (ROI) in phase 1. In the next phase deep learning algorithm is used to classify and predict the outcome. The object detection training is using the Haar Cascade Classifier which facilitates to obtain an accurate model. The model is trained by exploiting the open source tools namely Keras along with Tensorflow as backend. The hardware design includes Raspberry pi model 4, pi camera module V 1.3 and a buzzer. The video capturing, labeling of open or closed eyes and the score are obtained as a real time status with the buzzer output to alert the driver during drowsiness state. Training phase and Raspberry hardware design phase are deployed parallelly. The standalone hardware module is uploaded with the complete training AI software. Real time image acquisition by the design platform and predicting the driver drowsiness detection, depicts an accuracy of 95% to 96%. The activation function ReLU used during the prediction state in CNN improves the computational efficiency of the design model.