Abstract
Nowadays, automobile manufacturers make efforts to develop ways to make carsfully safe. Monitoring driver's actions by computer vision techniques to detectdriving mistakes in real-time and then planning for autonomous driving to avoidvehicle collisions is one of the most important issues that has beeninvestigated in the machine vision and Intelligent Transportation Systems(ITS). The main goal of this study is to prevent accidents caused by fatigue,drowsiness, and driver distraction. To avoid these incidents, this paperproposes an integrated safety system that continuously monitors the driver'sattention and vehicle surroundings, and finally decides whether the actualsteering control status is safe or not. For this purpose, we equipped anordinary car called FARAZ with a vision system consisting of four mountedcameras along with a universal car tool for communicating with surroundingfactory-installed sensors and other car systems, and sending commands toactuators. The proposed system leverages a scene understanding pipeline usingdeep convolutional encode-decoder network and a driver state detectionpipeline. We have been identifying and assessing domestic capabilities for thedevelopment of technologies specifically of the ordinary vehicles in order tomanufacture smart cars and eke providing an intelligent system to increasesafety and to assist the driver in various conditions/situations.