MCS-YOLO: AN AUTOMATED DRIVING ROAD ENVIRONMENT RECOGNITION SYSTEM WITH MULTISCALE OBJECT DETECTION

Abstract

Improving object identification algorithms' accuracy and speed is a critical issue in autonomous driving technology. Our MCS-YOLO procedure adds a direction consideration module to the backbone to further develop feature map spatial direction and cross-channel collection. Moreover, a multiscale little object detection structure builds aversion to thick minuscule articles, and CNNs utilize the Swin Transformer construction to focus on relevant spatial data. The BDD100K autonomous driving dataset shows that MCS-YOLO beats YOLOv5s in mean normal accuracy and recall rates. Curiously, our innovation detects 55 casings each second continuously driving circumstances. Further testing with YoloV5x6 shows a 0.798% mean typical accuracy increment. This exploration gives a solid and viable strategy for further developing autonomous driving object recognition, encouraging canny transportation frameworks