Mobile GPU based real-time status analysis and detection applications for driver assistant systems

With the development of computer vision technologies and the acceleration of deep learning studies, driver assistance systems have become quite widespread recently. These systems aim to detect critical situations by collecting the necessary data from the driver and vehicle environment in order to provide safety and driving convenience. Instant monitoring of the driver and the environment is important for the warning detection system. This will only be possible with the real-time operation of the system. In this thesis, a study has been carried out for advanced driver assistance systems (ADAS) that detects real-time in-vehicle and out-of-vehicle situations based on images with cameras connected to the embedded platform. Two applications focused on the driver and the environment, both inside and outside the vehicle, were carried out. While the developed system detects traffic signs, pedestrians and objects outside the vehicle, it provides warnings to the driver with fatigue and sleep detection by analyzing the driver's status inside the vehicle, monitoring phone and cigarette use and eye tracking. In the study, models were created with trainings on the graphics card (GPU) using the ready data set and the data sets specific to the study. In order to compare the detection rates, the system was tested in two low power and high performance embedded platforms (Jetson Xavier Nx, Nvidia Jetson Nano) and computer environment and the results were analyzed. As a result of the applications, a real-time ADAS prototype has been realized.