山区高速公路自动驾驶车道检测失效分析

Reliability Analysis of Autonomous Driving Lane Detection on Mountainous Freeways

摘要:

自动驾驶车辆在山区高速公路行驶时需要感知不断变化的道路线形、路段特征以及环境等因素，并解析安全行驶边界。针对山区高速公路，测试了基于激光雷达的感知性能，分析了车道线感知失效的影响因素。研究采用同济大学车载全息信息采集系统，实车测试路段为贵州省贵都高速公路，全程多次往返测试累计3 200 km。基于CatBoost模型和沙普利加和解释（SHAP）方法解释了11个特征变量对感知失效的影响。研究结果表明，存在以下特征时会检测出更高概率的感知失效，例如下游曲率最大值越大、坡度变化率越大、出口和入口路段、路面有纵向和横向减速标线等地面标线、天气不良、自车行驶速度越大等。该研究的结果不仅可以为山区高速公路提供更适合自动驾驶安全运营的措施，更好地适应自动驾驶车辆，而且还能为制定自动驾驶的运行设计域提供参考。

Abstract:

Autonomous vehicles need to perceive constantly changing road alignments， section characteristics， and environmental factors when driving on mountainous freeways， and analyze safe driving boundaries. This paper focuses on testing the lane detection performance of autonomous vehicles based on LiDAR in mountainous freeway environments and analyzes the influencing factors of lane perception failure. Using the Road and Traffic Holographic Data Acquisition System of Tongji University， the on-road vehicle test was conducted on the Guidu freeway in Guizhou Province， with multiple round trips totaling 3 200 kilometers. The impact of 11 feature variables on perceived failure was explained based on the CatBoost model and Shapley Additive exPlanations （SHAP）. The research results indicate that weather factors have the most significant impact on the perception failure of lane markings， followed by the influence of road alignments. Specifically， a higher probability of perceptual failure will be detected when there are the following characteristics： larger maximum curvature， larger grade change rate， exit， and entrance sections， longitudinal deceleration markings， lateral deceleration markings， higher driving speed， etc. The results of this paper can not only provide measures more suitable for the safe operation of autonomous driving on mountainous highways， adapting better to autonomous vehicles， but can also serve as a reference for establishing the operational design domain for autonomous driving.