A Profound Deep Learning approach for Driver Activity Recognition for Smart Vehicles

Abstract

Now a day’s making Journey had become an essential part of day to day life. And the safety becomes a major factor during travelling. Both self driving and paid driving results in unexpected accidents which sometimes leads to loss in life. Hence to ensure safety while driving driver choices and the decisions of drivers are fundamental variables that can influence safe driving. An  driver activity recognition model is proposed in which the experiment is done with the support of Convolution Neural Network (CNN).To conduct this experiment the input to this model are the raw images obtained from various activities of few drivers. Since all activities cannot be included , few and general activities like normal driving, checking left mirrors, checking right mirrors, verifying back mirror, switching indicators, drinking, messaging through mobile phone and attending calls are picked as input to the model in which first five activities are considered as primary and remaining three are secondary activities. These activities are captured using a camera fixed in the test area and images are collected from ten drivers. A Gaussian model GaM is used in handling the images like segmenting and makes it as input to the proposed model. To conduct this test a distinctive pre-prepared CNN model, AlexNet is used .The test resulted for the Eight activities was around of 81.8% accuracy when used with AlexNet.