Development of distance estimation system based on stereo vision to prevent vehicle-pedestrian accidents

Autonomous vehicles and driving support systems offer significant advantages in terms of traffic safety and vehicle driving convenience. With the development of new sensors and hardware technologies, faster and more powerful software methods, these intelligent vehicle systems are becoming more effective and safer every day. In this study, a stereo vision-based system is developed which identify ground-plane for driving, detect obstacles, calculate the distances of pedestrians and other objects in the driving region. Given the prevalence of camera systems, integration of these systems into vehicles offers an economical solution to other approaches (such as LIDAR). During the thesis study, new clustering and color segmentation algorithms have been developed and applied to the system to distinguish road, objects and pedestrians using color features of the image. The developed clustering algorithm distinguishes the data according to the distance and density properties without parameters and supervision, and this method is used to find color spaces on the image. In addition, the HOG filter is added to the system to determine the pedestrians around the vehicle. Besides, a new obstacle detection algorithm has been developed, and it has been possible to distinguish the obstacles on the road according to the characteristics such as color, depth and neighborhood with identifying the regions in the environment where the vehicle is going. Anthropometric proportions and spatial ranges depending on the distance are also taken into consideration when examining possible pedestrian zones. It is considered that the use of stereo camera based economical systems in driving support systems and autonomous vehicles will be widespread with the results produced in this thesis study. Increased driving support systems and autonomous vehicles will reduce vehicle-pedestrian accidents and will also reduce financial and moral losses.