Resilient object detection for autonomous vehicles: Integrating deep learning and sensor fusion in adverse conditions

Abstract

Autonomous vehicles (AVs) rely on advanced object detection systems to ensure safe navigation, especially under adverse weather conditions that can impair sensor visibility and introduce detection challenges. This manuscript provides a comprehensive analysis of state-of-the-art methodologies, focusing on deep learning frameworks, multi-sensor fusion techniques, and specialized datasets designed for AV object detection across various environmental conditions. We categorize approaches based on accuracy, computational efficiency, and resilience to challenging weather scenarios, offering insights into the strengths and limitations of each technique. Additionally, widely used datasets, such as KITTI and Waymo, along with synthetic and real-time datasets, are evaluated to assess their impact on detection accuracy in complex scenarios. While deep learning models demonstrate high accuracy, the integration of sensor fusion and transfer learning techniques further enhances robustness and adaptability. Our findings emphasize the importance of developing weather-resilient AV perception systems and provide recommendations for advancing object detection frameworks in autonomous driving applications.