Year: 2025
Type: article
Abstract: In the ever-evolving landscape of autonomous driving, object detection serves as the backbone of perception, dictating the safety and reliability of entire systems. Yet, navigating the complexities of real-world environments—ranging from adverse weather and occlusions to sensor noise and unknown objects—remains a formidable challenge. These obstacles underscore the urgent need to enhance the robustness of object detection systems, a cornerstone for the advancement of autonomous driving technologies. This survey delves into the latest research aimed at strengthening object detection robustness and explores critical aspects, such as advanced data augmentation methods, resilient model architectures, multi-modal feature representation, and emerging learning paradigms. Large-scale pre-trained models, comprehensive evaluation metrics, and testing protocols are also discussed to assess robustness under diverse conditions. By synthesizing existing research, this paper identifies current gaps and proposes pathways to balance performance and robustness while ensuring scalability. This work provides actionable insights for researchers and engineers, aiming to inspire the development of safer, more reliable, and adaptive object detection technologies for autonomous driving.
Author: Hengchen Cao
Institution: University of Edinburgh
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4407827913