A novel road attribute detection system for autonomous vehicles using sensor fusion

The development of Society of Automotive Engineers (SAE) Level 5 Autonomous Vehicles (AVs), which are capable of navigating a variety of roads and weather situations on their own, is examined in this study. Through the application of sophisticated computational algorithms, such as Improved You Only Look Once (YOLO) V5, Single Shot multibox Detector (SSD), Mask- Region based Convolutional Neural Networks (RCNN), and Nanodet, which are based on Convolutional Neural Networks (CNN), the research aims to enhance perception, prediction, and decision-making for secure and efficient autonomous navigation. The focus is on detecting road attributes like humps and potholes. Comparative analysis is carried out on both standard and custom datasets, leading to the selection of algorithms for real-time implementation. The proposed system employs a high-resolution camera mounted on a vehicle, connected to Graphics Processing Unit (GPU) accelerated embedded board, and implemented on the Robot Operating System (ROS) based software platform. Data collected on a specific route serves as valuable input for autonomous navigation. Additionally, the paper delves into the fusion of camera and Light Detection and Ranging (LiDAR) sensor data, introducing novel software architecture to seamlessly integrate road attribute detection into existing AV navigation pipelines on the ROS platform.