Multi-Class Driver Behavior Image Dataset

Distracted driving-related accidents are a critical global issue, especially as road traffic increases in densely populated areas. To address the challenge of driver distraction, we introduce a novel dataset that supports the development of real-time monitoring and detection systems by capturing authentic driver behaviors. Collected in Ashulia, Dhaka, Bangladesh, in October 2024, this dataset includes images captured under real-world driving conditions within both private vehicles and public buses. The photos were taken using personal mobile phones, ensuring a realistic and diverse set of visual data.

This dataset spans a wide range of driving behaviors, including safe driving, turning, texting, talking on the phone, and other potentially risky behaviors, such as drowsy driving. By depicting these behaviors in everyday driving scenarios, the dataset serves as a valuable resource for training and evaluating models designed to detect unsafe driving practices in real-time.The dataset includes high-resolution photos taken inside public buses and personal cars in Ashulia, Dhaka, Bangladesh, under actual driving circumstances. The photographs, which were taken using the cameras on personal cell phones, offer a genuine and varied collection of visual information under normal driving circumstances. The following five behavioral classes comprise the dataset:
I. Safe Driving: Images showing a driver who seems to be paying attention to the road, both hands on the wheel, and concentrated or 1 hand on the steering wheel and other on the gear stick. This is the perfect example of driving without distractions.
II. Turning: Photographs that show drivers changing direction during turns by moving their heads or full bodies. This behavior is crucial for figuring out how focused the driver is on everyday tasks like rotating the steering wheel.
III. Texting Phone: Pictures of drivers using their phones, whether it is to type messages or to interact with the screen. Since texting and driving is one of the main causes of distracted driving, this training is very important for identifying it.
IV. Talking Phones: When drivers talk on their phones or hold them up to their ears while driving a vehicle. This category aids in identifying actions connected to phone talks, which are another frequent source of interruptions.
V. Others: Contains any actions that go against safe driving practices, like drinking water or anything while driving, sleeping while driving, or talking with someone behind while driving.
Relevant photos are included in each session, and they differ in terms of vehicle type and illumination to represent the variety of driving situations found in the real world. Because the images are unprocessed and unannotated, there is freedom in how machine learning applications pre-process them.

与分心驾驶（distracted driving）相关的交通事故是一项严峻的全球性挑战，尤其是在人口稠密地区道路交通流量持续攀升的背景下。为应对驾驶员分心带来的挑战，我们推出了一款全新数据集，通过采集真实的驾驶员行为数据，为实时监测与检测系统的研发提供支撑。
该数据集于2024年10月在孟加拉国达卡市阿舒利亚（Ashulia）采集，涵盖了私家车与公共巴士内真实驾驶场景下的图像数据。所有照片均通过个人手机拍摄，确保了视觉数据的真实性与多样性。
本数据集覆盖了多种驾驶行为类别，包括安全驾驶、转向操作、手机发短信、手机通话，以及其他潜在危险行为（例如疲劳驾驶（drowsy driving））。通过还原日常驾驶场景中的各类行为，该数据集可作为训练与评估实时不安全驾驶行为检测模型的宝贵资源。
本数据集包含孟加拉国达卡市阿舒利亚地区真实驾驶场景下的高分辨率车内图像，拍摄载体为个人手机，涵盖私家车与公共巴士两类车型，能够提供真实且丰富的视觉信息。
本数据集包含以下五类驾驶行为类别：
一、 安全驾驶：指驾驶员专注路面、双手握方向盘，或单手握方向盘、另一只手操作换挡杆的图像场景，属于无分心的标准驾驶行为范例。
二、 转向操作：指驾驶员通过转动头部或调整全身姿态完成车辆转向的画面，该类别可用于分析驾驶员在日常转向操作中的注意力集中度。
三、 手机发短信：指驾驶员在驾驶过程中使用手机编辑短信或操作手机屏幕的画面。由于驾驶时发短信是分心驾驶的主要诱因之一，该类别对分心驾驶检测模型的训练至关重要。
四、 手机通话：指驾驶员在驾驶时将手机举至耳边通话的行为画面，该类别可用于识别另一类常见的驾驶分心行为。
五、 其他违规行为：涵盖所有违反安全驾驶规范的行为，例如驾驶时饮水、疲劳驾驶，以及与后排乘客交谈等。
每一组采集会话均包含对应场景的相关图像，且涵盖不同车型与光照条件，以还原真实世界中多样化的驾驶场景。由于图像未经预处理与标注，可为机器学习应用的预处理环节提供充分的自定义空间。