New road and sign recognition methods for driver assistance systems

The interest in intelligent transportation systems is increasing day by day and therefore many driver support systems are being developed. Especially, systems that use images have been given importance recently. Research on recognition and understanding of road environments such as traffic signs and lane markings is examined. Within the scope of the thesis study, methods have been proposed for traffic signs detection and classification, traffic light classification, and road lane line detection. Additionally, a data set consisting of traffic signs in Turkey is produced and tested at the classification stage. Image processing and deep learning methods are used in the operations. Regarding traffic sign detection, the images taken from inside the vehicle are processed using the recommended method based on color and shape. Multiple studies have been conducted for traffic sign classification, including transfer learning, color space-based, and ensemble learning, and the best performance is achieved with ensemble learning. A convolutional neural network-based model is proposed for traffic light classification. Different experiments are carried out for model parameter selection, the results are compared and the most suitable parameter values for the model are obtained. Finally, a lane line detection method with image processing and Support Vector Regression curve fitting is proposed.