Driver Anomaly Detection (DAD) dataset
收藏资源简介:
Driver Anomaly Detection (DAD)数据集是由慕尼黑工业大学创建的一个大规模、多视角、多模态的视频基准数据集,旨在通过深度学习方法监测驾驶员行为异常。该数据集包含正常驾驶视频及一系列异常驾驶行为,支持深度卷积神经网络(CNN)的训练。数据集采用前视和顶视两种视角,以及深度和红外两种模态,确保在不同光照条件下系统的可操作性。DAD数据集特别适用于开放集识别场景,即在测试时可能出现未知行为,这对于提高驾驶员监控系统的鲁棒性至关重要。
The Driver Anomaly Detection (DAD) dataset is a large-scale, multi-view, multi-modal video benchmark dataset developed by the Technical University of Munich. It is designed to detect abnormal driver behaviors using deep learning approaches. This dataset contains both normal driving videos and a range of abnormal driving behaviors, supporting the training of deep convolutional neural networks (CNNs). The dataset adopts two perspectives (front-view and top-view) and two modalities (depth and infrared), ensuring the system's operability across varying lighting conditions. The DAD dataset is particularly well-suited for open-set recognition scenarios, where unknown driving behaviors may emerge during testing, which is critical for enhancing the robustness of driver monitoring systems.



