synthetic-driver-monitoring-detection
收藏资源简介:
Synthetic DMS(驾驶员监控系统)数据集是一个由AnywayLabs.ai生成的完全合成图像数据集,专门用于驾驶员监控和分心检测任务。该数据集旨在为汽车环境中的驾驶员风险行为检测模型提供训练数据,采用有监督目标检测的范式。数据集包含1,356张带标注的训练图像,覆盖四种驾驶员分心行为类别(如饮水、打哈欠、打电话、发短信等),并提供了多个摄像头视角(FOV1, FOV2, FOV3及其变体)。图像分辨率为1376×768,标注采用YOLO格式(归一化边界框)。数据集通过专有的合成生成框架创建,能够对行为类型、驾驶员姿势、遮挡情况、光照条件和车内背景等进行可控变化,旨在扩展行为分布、引入受控的外观变化,从而提高模型对未见真实世界驾驶员行为的鲁棒性。该数据集适用于驾驶员分心检测模型训练、在真实舱内数据微调前的合成预训练、汽车安全系统开发以及多视角行为识别研究。与真实世界驾驶员监控数据集相比,本数据集无隐私顾虑、标注成本低、行为覆盖可配置且类别平衡可控。
The Synthetic DMS (Driver Monitoring System) dataset is a fully synthetic image dataset generated by AnywayLabs.ai, specifically designed for driver monitoring and distraction detection tasks. It aims to provide training data for models detecting driver risk behaviors in automotive environments, following a supervised object detection paradigm. The dataset includes 1,356 annotated training images covering four driver distraction behavior categories (such as drinking, yawning, calling, texting, etc.), and offers multiple camera perspectives (FOV1, FOV2, FOV3, and their variants). The image resolution is 1376×768, with annotations in YOLO format (normalized bounding boxes). Created through a proprietary synthetic generation framework, it allows controlled variations in behavior types, driver postures, occlusion conditions, lighting conditions, and in-car backgrounds, aiming to expand behavior distributions and introduce controlled appearance variations, thereby enhancing model robustness to unseen real-world driver behaviors. This dataset is suitable for training driver distraction detection models, synthetic pre-training before fine-tuning on real in-cabin data, automotive safety system development, and multi-view behavior recognition research. Compared to real-world driver monitoring datasets, it has no privacy concerns, low annotation costs, configurable behavior coverage, and controllable class balance.




