Pose Variation and Occlusion based Driver expression recognition using Convolutional Neural Network

Abstract To enhance driver safety, various technological developments have been made in vehicle systems from the past decade, accessible to the drivers inside the vehicle to track driver's emotions before causing accidents. These systems assist "ADAS" (Advance Driver Assistance Systems), which can help to improve safety and provide enough potential for the driver to react early before accidents. It is the primary factor that influences the popularity and safety of vehicles. The lack of monitoring of the driver emotion is causing to increase in the number of accidents per year. Tracking the driver's emotion in various angles in different illumination conditions is much needed to predict the correct emotion or behavior of the driver. Occlusions like hair and sunglasses are also significant factors influencing the drivers to make accidents and their emotions. To prevent these, two novel approaches are proposed to detect the occlusions. The pose variations involving driver emotion in which seven classes of expressions are being detected from the driver in different illumination conditions. We achieved better accuracies when compared to state-of-the-art existing methods.