Modelling of environment-interactive safe drive assistant systems

In this thesis, new approaches related to Advanced Driver Assistant Systems have been developed. Autonomous vehicles are intelligent-vehicles that can act driverless. The reasons such as violation of the traffic rules, inattention, fatigue, and insomnia are all significant obstructions to achieve full safety in cars with a driver. In contrast, thanks to the autonomous vehicles that are not in control of the driver are aimed to eliminate the problems caused by driver. Therefore, in this study as an ideal model of driver behavior, full safe driving has been adopted to fully autonomous driving model. Intelligent systems referring Driver Assistance Systems and active safety have been becoming as prominent terms for implementation of autonomous-driving. In this study, new approaches related to Driver Assistant Systems have been developed to contribute the concept of autonomous-driving. In this work, owing to the aim of safe driving, some approaches related to high level driver assistant systems have been focused, as exemplifying the risk estimate, driver and road profile detection, and accident prediction via risk estimation. In this sense, this thesis includes specified studies hat are approaching vehicle detection and tracing, overtaking risk analysis, a feature selection method for both Aggressive/Calm driver detection problem and road slope type detection problem, risk estimation to get safe cornering, and traffic scene analysis for most risky region detection. Thanks to new approaches developed in the studies, we have aimed to tackle with the problem, without the need of the external hardware components already existing in a vehicle such as front and rear camera, gradually becoming standard in cars, or CAN-bus data involving internal informations belonging to the car. In this thesis; the approaches such as Optical-edge flow method for detection of object motion and a new clustering algorithm; overtaken car based upon an approach of risk estimation, a new feature selection for detection of driver mood and road slope type problems; geometric approaches such as road curvature, bend slope type, and bend direction estimate for get cornering; and risk analysis using Graph-Cut method for detection of most risky region in the urban traffic scene have been developed.