基于深度学习的复杂天气自动驾驶障碍物检测研究

Research on Autonomous Driving Obstacle Detection in Complex Weather Based on Deep Learning

摘要: 在自动驾驶领域，精准识别道路障碍物是保障行车安全与实现智能规划的核心技术基础。然而，在夜间低照度、雨雾等复杂天气条件下，道路场景可见性显著下降，目标边缘模糊、尺度变化剧烈，传统目标检测算法在实际应用中常面临漏检和误检的挑战，难以满足系统对可靠性的要求。为此，本研究聚焦于复杂天气环境下的自动驾驶障碍物识别问题，引入了一种基于YOLOv8改进的目标检测模型。首先，分析了复杂天气与光照变化对自动驾驶视觉感知的影响，在此基础上构建了包含正常天气、夜间、恶劣天气及夜间叠加恶劣天气等多种典型场景的数据集。其次，以YOLOv8为基线模型，从特征表达能力与训练稳定性角度对网络结构和训练策略进行改进，以提升模型在复杂场景中对低可见度目标的辨识能力。最终，通过多组对比实验，系统性地评估了改进模型在不同工况下的检测性能。实验结果表明，与基线模型相比，所提出的方法在复杂天气场景下取得了更优的检测效果，尤其在夜间与恶劣天气等复杂气象条件下，模型整体性能及环境适应能力均获得显著改善。研究结果表明，该方法能够有效增强自动驾驶系统在复杂环境下的障碍物检测能力，为实际道路场景中的自动驾驶感知提供了有力支持。

Abstract: In the field of autonomous driving, the accurate identification of road obstacles is a foundational technology for ensuring driving safety and enabling intelligent planning. However, under complex weather conditions such as low illumination at night, rain, and fog, the visibility of road scenes significantly decreases, with blurred object edges and drastic scale variations. Traditional object detection algorithms often face challenges of missed and false detections in practical applications, making it difficult to meet the system’s reliability requirements. To address this, this study focuses on the obstacle recognition problem for autonomous driving in complex weather environments and introduces an improved object detection model based on YOLOv8. First, the impact of complex weather and illumination changes on visual perception in autonomous driving was analyzed. Based on this, a dataset was constructed containing various typical scenarios, including normal weather, nighttime, adverse weather, and combined nighttime with adverse weather conditions. Second, using YOLOv8 as the baseline model, improvements were made to the network structure and training strategies from the perspectives of feature representation capability and training stability, aiming to enhance the model’s ability to recognize low-visibility targets in complex scenes. Finally, through multiple sets of comparative experiments, the detection performance of the improved model under different operating conditions was systematically evaluated. The experimental results show that compared to the baseline model, the proposed method achieves superior detection performance in complex weather scenarios. Particularly under challenging meteorological conditions such as nighttime and adverse weather, both the overall performance and environmental adaptability of the model are significantly improved. The findings demonstrate that this method can effectively enhance the obstacle detection capability of autonomous driving systems in complex environments, providing robust support for autonomous driving perception in real-world road scenarios.