Semantic road segmentation based on adapted Poly-YOLO

Abstract With artificial intelligence continuing to change people’s everyday life in profound ways, the desire endow vehicles with ability drive autonomously has emerged for years. Thus, autonomous driving become a popular field. The task can be divided into three general procedures: perception, planning, and locomotion. first foremost part of these procedures is perception task. Among those methods, most prevailing one semantic segmentation, which annotating predicting object located at pixel level, meaning nearly all pixels should classified certain categories. However, this method provides enough accuracy while bringing considerable computational burden. implementing real-time road segmentation on still costly In paper, an adapted model improved upon Poly-YOLO baseline proposed, well-developed detection algorithm providing bounding polygons enclose target object, forming polygon mask similar that segmentation. This paper endeavors enhance model’s detecting variously sized targets greatly fine-tune generate more proximate enclosing polygons. experienced leap performance compared model.