A Multi-Modal Framework For Road Attribute Detection And Evasive Maneuvering In Autonomous Vehicles
Keywords:
Object Detection, Classification Models, YOLOv4, YOLOv6, YOLOv8, VGG-16, ResNet, Road Attribute Detection, Pothole Detection, Hump Detection, Autonomous Vehicles, architectural Innovations, Detection Accuracy, Computational Efficiency, Real-world Applicability, Depth Estimation, Solid State LiDAR.Abstract
This paper presents a robust multi-modal framework for the real-time detection of road anomalies and subsequent autonomous vehicle response. This paper presents a comprehensive comparison of road attribute detection techniques using advanced machine learning models YOLO v4, v6, v8, VGG-16 and ResNet alongside an analysis of depth estimation for potholes using LiDAR data. The objective is to evaluate the effectiveness of these models in accurately detecting road anomalies such as potholes and humps, and to understand how such detections can influence the response mechanisms of an autonomous vehicle. The models were rigorously tested under controlled simulations to detect various road attributes. YOLO versions, renowned for their object detection capabilities, were compared against the classification-oriented VGG-16 and ResNet-34 models to assess their precision, speed, and utility in real-time scenarios. Depth estimation was performed using LiDAR data to gauge the severity of detected potholes, enhancing the vehicle’s decision-making process regarding speed adjustment and maneuvering. Fusion between 2D data using computer vision architectures like YOLOv8, YOLOv4, VGG-16, and ResNet-34 with 1D LiDAR depth profiling, the system identifies and evaluates the severity of potholes and humps. The findings reveal distinct strengths and limitations of each model in the context of autonomous driving. The paper also explores how the detection information is utilized by an autonomous vehicle’s control system to execute responsive actions such as deviation from potholes and speed reduction over humps. This integration of detection and responsive action aims to improve navigation safety and passenger comfort, showcasing the potential of combining multiple sensory data and machine learning models in the development of autonomous vehicle technologies. The results contribute valuable insights into the selection of appropriate models and systems for enhanced road attribute detection and response strategies in the evolving landscape of autonomous driving.
