Autonomous vehicle application suitable for traffic conditions using deep learning

In a period when technology develops and firms increase their investments in the effort of globalization, it has become inevitable for new ideas to emerge and competition to increase. As a result of the increase in the hardware capacities of the devices over time, autonomous system studies have accelerated in many sectors. Advances in machine learning and deep learning have enabled the solution of complex problems and the development of many different applications. Autonomous vehicles are vehicles that do not require human intervention and have the ability to move on their own. In recent years, autonomous vehicles have gained a place in both the academic field, the defense industry and the private sector, including military, commercial and research projects. It is predicted that with the development and widespread use of autonomous vehicles, driver-related traffic accidents can be prevented, fuel savings will be achieved and traffic congestion will be eliminated. In this context, the basic working principles, functions and variables of artificial neural networks were examined in the thesis study. Then, a convolutional neural network model was created based on the MNIST dataset consisting of numbers from 0 to 9. While creating the model, all layers were examined in detail and the success results of the model were evaluated with graphics. Finally, a deep learning model was created for the autonomous movement of a driverless vehicle and neural network training was carried out. During the study, the Torch library was used and a deep learning application running on the GPU was implemented. Within the scope of the study, a car kit, Jetson Nano development board, Raspberry Pi camera modüle and ultrasonic distance sensor were used, and a track was prepared for autonomous driving. With the trained neural network model, the vehicle was provided to complete the track autonomously and the results of the study were presented with graphics.