A Novel Solution to the Real-Time Lane Detection and Tracking Problem for Autonomous Vehicles by Using Faster R-CNN and Mask R-CNN

Autonomous vehicle technology has advanced in the automobile sector. Autonomous vehicle technology aims to make driving safer and reduce driver-caused traffic accidents. Autonomous driving technology work toward this. Lane detection and tracking are crucial to autonomous driving systems. Mostly the image processing techniques are mainly utilized for the lane detection in the literature. But, while performing image processing techniques for lane detection and tracking, two basic problems are mainly encountered. First one is image also needs to work with a specific area on the image to reduce the processing load and to work for the correct area. The region of interest (ROI) process is often used to filter the area to be worked from the image. However, since fixed coordinates are provided for this operation, the vehicle restricts the oper-ation of the vehicle in areas where it must be rotated. Second one is the weather conditions are very effective in the detection of lanes by utilizing image processing techniques. There are serious problems with image processing and detection from cloudy, sunny or momentary changes in the air. This study uses deep learning methods against these two basic problems. Using the Mask R-CNN and faster R-CNN algorithms together, these two basic problems for lane detection and tracking are eliminated and successfully implemented. The problem solved by two algorithms has been tested experimentally on a developed tool. Both the originally developed da-taset and the KITTI dataset were used separately in the model training carried out for the experimental tests. Both systems work well for lane detection and tracking, according to tests.