Zhou Quanxing, Carlos Babaran Jr
ABSTRACT
The automotive driving management system represents a cutting-edge domain within the automotive industry, where the performance of its core object detection system is crucial in determining the safety and intelligence of autonomous driving. This paper investigates object detection technology for complex driving environments. Despite significant advancements driven by deep learning, current algorithms still face challenges in accuracy, robustness, and real-time processing, particularly under high-density, dynamically changing, and extreme weather conditions. To address these issues, this research systematically optimizes and innovatively enhances advanced object detection algorithms by introducing attention mechanisms, multi-scale feature fusion, and illumination adaptation strategies tailored for high dynamic range scenes. Additionally, by employing refined anchor matching and deep learning optimization, the model’s detection performance and generalization for small and distant objects, as well as challenging conditions such as backlighting and nighttime scenarios, were substantially improved. The study also designed a user-friendly interactive interface using PySide6, reducing operational barriers and facilitating the practical adoption of the technology. Evaluation on public benchmark datasets and under various complex environmental conditions demonstrates that the proposed optimized algorithm significantly improves key performance metrics, including precision and recall, while maintaining high real-time processing efficiency and exhibiting exceptional robustness in extreme scenarios.
Keywords: Autonomous Driving, Object Detection, Robustness, Human-Computer Interaction (HCI) Interface
https://doi.org/10.57180/sanb8246