Determination of highway type, intersections and curves for intelligent transportation systems by deep learning methods

Recently, new ideas and scenarios have been produced for autonomous vehicles and advanced driver assistance systems developed in the field of intelligent transportation systems. In autonomous vehicles or driver assistance systems, computer vision techniques are often used for road recognition, or laser signal processing techniques are applied. Image processing methods can produce lower cost solutions than laser signal processing methods. In addition to image processing techniques, the object and path type can be determined using deep learning methods. In this thesis, real images of important regions such as intersections, pedestrian crossings, bends, left-right intersections on the road have been dealt with for the first time as a recognition problem and determined by image processing and deep learning method. In the image processing stage, a method has been developed for the determination of the relevant region by clipping the top and bottom points according to the structure and attraction of road images and the acquired images were recognized with deep learning architecture. Images have been tested with the method we developed and different recognition algorithms. It has been found that our model gives very good results both according to algorithms and studies developed using different datasets in the literature. Using camera images, real-time systems can be prepared and cost-effective solutions can be provided. It can be said that this developed method is applicable to driver support systems and it is an effective structure that can be used in many fields such as warning of vehicles and drivers. Also, this study can be used for detection and verification systems as well as whether traffic signs on highways match actual road conditions.