用卡车CAN信号预测超车行为，提升ADAS安全决策能力
Overtake Detection in Trucks Using CAN Bus Signals: A Comparative Study of Machine Learning Methods
- 基于五辆实车的CAN数据，对比神经网络、随机森林与SVM分类效果
- 融合多车数据并采用评分级融合策略，实现93%真负率与86.5%真正率
- 适用于车载AI系统研发，尤其关注数字后视镜场景下的驾驶行为预测
卡车安全超车对防止事故和保障交通效率至关重要。准确预测此类操作对高级驾驶辅助系统（ADAS）及时决策极为关键。本研究利用沃尔沃集团提供的五辆在役卡车的控制器局域网（CAN）数据，评估人工神经网络（ANN）、随机森林（RF）和支持向量机（SVM）三种常见分类器在车辆变道检测中的表现，并分析不同预处理配置对性能的影响。研究发现，交通条件差异显著影响信号模式，尤其是非超车类样本，若训练数据缺乏多样性则会降低分类性能。由于数据采集于真实开放环境，类别分布无法预先保证。但使用多车数据训练可增强泛化能力，减少特定条件偏差。每辆车独立分析显示，分类准确率（尤其是超车识别）依赖于每车的训练数据量。为此，采用评分级融合策略，在多数情况下取得最优单车性能。整体融合结果达到真负率（TNR）93%、真正率（TPR）86.5%。该研究是BIG FUN项目的一部分，探索人工智能如何应用于车辆日志数据以理解并预测驾驶行为，尤其关注作为传统外后视镜数字替代品的相机监控系统（CMS）。
原文摘要 · Abstract (English)
Safe overtaking manoeuvres in trucks are vital for preventing accidents and ensuring efficient traffic flow. Accurate prediction of such manoeuvres is essential for Advanced Driver Assistance Systems (ADAS) to make timely and informed decisions. In this study, we focus on overtake detection using Controller Area Network (CAN) bus data collected from five in-service trucks provided by the Volvo Group. We evaluate three common classifiers for vehicle manoeuvre detection, Artificial Neural Networks (ANN), Random Forest (RF), and Support Vector Machines (SVM), and analyse how different preprocessing configurations affect performance. We find that variability in traffic conditions strongly influences the signal patterns, particularly in the no-overtake class, affecting classification performance if training data lacks adequate diversity. Since the data were collected under unconstrained, real-world conditions, class diversity cannot be guaranteed a priori. However, training with data from multiple vehicles improves generalisation and reduces condition-specific bias. Our pertruck analysis also reveals that classification accuracy, especially for overtakes, depends on the amount of training data per vehicle. To address this, we apply a score-level fusion strategy, which yields the best per-truck performance across most cases. Overall, we achieve an accuracy via fusion of TNR=93% (True Negative Rate) and TPR=86.5% (True Positive Rate). This research has been part of the BIG FUN project, which explores how Artificial Intelligence can be applied to logged vehicle data to understand and predict driver behaviour, particularly in relation to Camera Monitor Systems (CMS), being introduced as digital replacements for traditional exterior mirrors.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。