Distracted Driver Detection using Inception V1

Abstract

A major contributing factor in car accidents is driver distraction. This research suggests a distraction detecting system for drivers that detects various forms of distractions by watching the driver with a camera in an effort to decrease traffic accidents and enhance transportation safety. To develop practical driving situations and to test the algorithms for distracted detection, an assisted driving testbed is being constructed. Pictures of the drivers in both their regular and distracted driving postures were taken for the authors’ dataset. The VGG-16, AlexNet, GoogleNet, and residual network are four deep convolutional neural networks that are developed and assessed on a platform with integrated graphics processing units. A voice warning system is developed to notify the driver when they are not paying attention to the road. As VGG-16 is a huge network, it takes more time to train its parameters. On the other hand, ‘texting left’ was misclassified with ‘safe driving’ in some scenarios when the steering wheel blocked the left hand. According to experimental findings, the proposed strategy works better than the baseline approach, which only uses 256 neurons in the fully linked layers. GoogleNet uses inception module, used for running multiple operations (pooling, convolution) with multiple filter sizes in parallel so that it is not necessary to face any trade-off. It takes less time to train its parameters.