Detection of Driver Behaviours using Convolutional Neural Network with Modified Inception Module

Drivers who are distracted are the significant causes of traffic accidents. Traffic accidents could be minimized with a system that gives an early warning about their conditions. This study aims to develop detection systems for driver conditions. The proposed method consists of a conventional convolutional layer followed by two inception modules with each depth of two layers, comprising five layers. The Inception module is modified by adding larger $7\times7$ kernels. We collect a dataset from four users, each comprising ten driver conditions. Our experiments for user-dependent systems, i.e., conditions where all users are used in training and testing data, show an accuracy of 99.33%. However, the performance drops for user-independent systems, i.e., conditions where users for training are different from that of the testing. Nevertheless, our systems perform better than a conventional CNN in both evaluation conditions.