恶劣环境图像数据集

<p>该数据集包含超过178000个高分辨率立体图像对，分为 27 个序列，反映了雪、雨、雾和弱光等丰富多样的条件。它涵盖了驾驶场景和环境背景的动态变化，包括大学校园、住宅街道和城市环境。Stereo Image Dataset (SID)恶劣环境图像数据集</p><p>数据来源：公开网站<a href="https://deepblue.lib.umich.edu/data/concern/data_sets/cc08hg37c" rel="noopener noreferrer" target="_blank">https://deepblue.lib.umich.edu/data/concern/data_sets/cc08hg37c</a></p><p>数据规模：超过 178000个高分辨率立体图像对</p><p>数据特点：反映了雪、雨、雾和弱光等丰富多样的驾驶条件。</p><p>SID 数据集是为了支持自动驾驶系统的先进研究而精心策划的，特别关注在恶劣天气和光照条件下的感知能力。该数据集包含超过 178k 对高分辨率立体图像，分为 27 个序列，反映了包括雪、雨、雾和低光照在内的丰富多样的条件。它涵盖了驾驶场景和环境背景的动态变化，包括大学校园、住宅街道和城市环境。该数据集旨在通过部分遮挡的相机镜头和不同的可见性等场景来挑战感知算法，促进鲁棒性计算机视觉模型的发展。访问图像数据无需专用软件或脚本，因为文件以标准 PNG 格式提供。然而，研究人员和开发者可能需要使用他们的图像处理和计算机视觉工具包，才能有效地在他们的工作中利用该数据集。</p><p>立体图像数据集（SID）使用安装在车辆上的 ZED 立体相机录制，在不同天气条件和一天中的不同时间以 720p 分辨率、20Hz 帧率拍摄图像。在暴雨期间，相机偶尔会被放置在挡风玻璃下方以防止直接暴露在雨滴中。固定路线覆盖了密歇根大学-迪尔伯恩校园以及其他城市和住宅区域，提供了多样化的环境。相机系统生成的立体图像对具有工厂校准，用于深度估计。使用 ZED 相机 API 创建了自定义数据记录软件，最初将图像存储为 BMP 格式，后来转换为 PNG 格式以实现高效数据存储。元数据注释，包括天气条件、一天中的时间以及道路状况，在拍摄后手动验证和记录，为每个图像序列提供详细背景信息。</p><p>应用场景：用于自动驾驶中恶劣环境场景</p>This is the Stereo Image Dataset (SID), a harsh-environment image dataset. It contains over 178,000 high-resolution stereo image pairs, divided into 27 sequences, covering diverse conditions such as snow, rain, fog, and low light. It encompasses dynamic changes in driving scenarios and environmental backgrounds, including university campuses, residential streets, and urban environments. Data Source: The publicly available dataset can be accessed at https://deepblue.lib.umich.edu/data/concern/data_sets/cc08hg37c Data Scale: Over 178,000 high-resolution stereo image pairs Data Characteristics: Reflects a wide range of driving conditions including snow, rain, fog, and low light. The SID dataset was carefully curated to support advanced research on autonomous driving systems, with a particular focus on perception capabilities under adverse weather and lighting conditions. It contains over 178k high-resolution stereo image pairs, divided into 27 sequences, covering diverse conditions including snow, rain, fog, and low light. It encompasses dynamic changes in driving scenarios and environmental backgrounds, including university campuses, residential streets, and urban environments. This dataset is designed to challenge perception algorithms with scenarios such as partially occluded camera lenses and varying visibility, promoting the development of robust computer vision models. No specialized software or scripts are required to access the image data, as the files are provided in standard PNG format. However, researchers and developers may need to use their image processing and computer vision toolkits to effectively utilize this dataset in their work. The Stereo Image Dataset (SID) was recorded using a ZED stereo camera mounted on a vehicle, with images captured at 720p resolution and 20Hz frame rate across different weather conditions and times of day. During heavy rain, the camera was occasionally placed under the windshield to prevent direct exposure to raindrops. Fixed driving routes covered the University of Michigan-Dearborn campus as well as other urban and residential areas, providing diverse environments. The stereo image pairs generated by the camera system have factory calibration for depth estimation. A custom data logging software was created using the ZED Camera API, which initially stored images in BMP format before being converted to PNG format for efficient data storage. Metadata annotations, including weather conditions, time of day, and road conditions, were manually verified and recorded after image capture, providing detailed background information for each image sequence. Application Scenarios: Used for harsh environment scenarios in autonomous driving.