Drivable Road Area Detection with Regression Output CNN

Abstract

Nowadays, many methods are developed on autonomous vehicles and driver assistance systems to prevent traffic accidents and support drivers. In this work, a drivable area detection method based on CNN and regression is proposed. In the proposed method, Cityscapes dataset, which is open to sharing on the Internet is used as dataset. The images in the dataset are cut into slices to obtain new input images. With those images, a CNN based deep learning network is trained. By applying linear regression on the characteristics of the output of the network, the road boundary points in the relevant slice are fried to be determined. Experimental results have shown that the developed method has real-time operating performance and the results can be improved.