Abstract

Road detection for autonomous navigation is an important task in computer vision and image processing with the aim of identifying and segmenting navigable areas from road image frames. Edge detection, which identifies boundaries of objects in an image, plays a foundational role in the process and is a widely used technique in many cases, such as object recognition and tracking, image segmentation and enhancement, and satellite image analysis, to name a few. There are several edge detection methods, such as Roberts, Sobel, Canny, and Prewitt. In this research, two prominent edge detectors, Sobel and Canny detectors, are analyzed and compared in the drivable region task. The two edge detector models are applied to a stream of real-life road images obtained using the Huawei Lite 40 JNY-LX1 RGB sensor camera and secondary road image frames (BDD Dataset) obtained from Kaggle.  Drivable regions were detected using a binary edge mask and identifying the largest connected component in the road image with a morphological closing operation. Each edge detector’s performance for the drivable detection task was compared and evaluated using Pixel Accuracy (PA) and Intersection over Union (IoU). In a series of experiments conducted, the results show that good drivable regions, which facilitate safe autonomous navigation, can be detected with Canny showing superior performance.