Applied Computer Vision on Advanced Driving Systems

Authors

Rodriguez Zepeda, Abraham

Castro Castro, Rigoberto

Reyes Duke, Alicia Maria

Abstract

As driving systems, and autonomous vehicles are developing, computer vision plays a vital role on the perception of these systems, this allows vehicles to detect the drivable area, or the area between lanes, and detect objects, with the purpose of reacting to its environment, generating a certain level of autonomy. In this paper, an algorithm based on computer vision techniques such as sliding windows, color thresholding, edge detection, and perspective transformation are used for lane detection, once the lanes are detected the left and right curves can be fitted with polynomial regression, obtaining the drivable area, which is the area between the detected curves. For object detection, a pretrained version of YOLOv5 is implemented. The algorithm was implemented using a custom dataset generated on various sectors of San Pedro Sula, Honduras, showing promising results on different scenarios. For a dataset of 1705 processed images, the lane detection accuracy of the algorithm under ideal conditions is 94%.