Automated Overtake Assist System Using YOLOv4 and Mask R-CNN for Enhanced Autonomous Driving

The advent of computer vision technology has significantly advanced the development of automated driving systems, enabling more sophisticated and safer driving aids. This paper presents an Automated Overtake Assist (AOA) system that leverages state-of-the-art computer vision algorithms - YOLOv4 and Mask R-CNN - for lane and vehicle detection. Additionally, an ultrasonic sensor is employed to accurately measure the distance to the vehicle directly in front. The system is implemented on a Raspberry Pi microcontroller, integrated with a high-definition webcam to capture real-time visual data, facilitating intelligent decision-making and enhancing the safety and efficiency of overtaking maneuvers in Autonomous vehicles(AVs). The error rate of the system is notably smoother than typical human error, ranging between approximately 0.1% and 0.25%. Training yielded a total loss of 0.5702 (0.6646), with loss components including: lossclassifier at 0.0922 (0.1262), loss box at 0.1815 (0.2140), lossmask at 0.1725 (0.1695), lossobjectness at 0.0209 (0.0226), and loss rpn at 0.0748 (0.1323). Consequently, the system achieves an accuracy of approximately 98%. Our results demonstrate the AOA system's capability to perform accurate lane detection, vehicle recognition, and distance measurement, ensuring a seamless and safe overtaking process in response to the driver's intent.