Ship Hull Vinyl Dataset

An inside sweet water tank at the Institute for Maritime Technology (IMT) in Cape Town was used to take a controlled underwater dataset using a prototype sensor rig - 24 October 2023. The target scene is a life-sized photomosaic of the side of a commercial ship hull that IMT had previously taken when the ship was last in dry dock. The photomosaic is printed onto a large vinyl tarpaulin and hung along the side of the tank.Because the tank platform is 1m from the surface of the water, two 3m PVC pipes were used to manoeuvre the rig through the water, seen below. The rig is moved horizontally along the side of the platform facing the vinyl surface. Marks were placed on the PVC pipes as a visual aid to keep the rig at a relatively constant depth when moving horizontally. Once at the end of the vinyl, the rig is lowered directly downward to the next marked depth and moved horizontally back along the platform, creating a square ‘C’ shape. The data collection starts at 0.5m depth and returns at 1m depth.We provide the data in the form of ROS bags. The data is supplied in two formats: raw and pre-processed. The raw data has not been edited. The pre-processed data has been synchronized and images have been stereo rectified and undistorted. With the ROS bags we also include the stereo camera and stereo-inertial (camera-IMU) calibration output files in yaml format.The original calibration data for camera and stereo-inertial calibrations are also included in ROS bag format so that users can perform their own calibrations if desired. On the sister GitHub, in the Meta Data folder, we detail the data collection method in a PDF for contextual understanding of the dataset and provide the data sheets of the sensors used along with supporting code.

在开普敦的海洋技术研究所（IMT）内部的一座淡水储罐中，利用原型传感器装置，于2023年10月24日进行了受控水下数据集的采集。目标场景为IMT在船舶上次进干坞时拍摄的商业船体侧面的一幅等比例摄影镶嵌图。该摄影镶嵌图被印制在一张大型的乙烯基防水布上，并悬挂在储罐的侧面。由于储罐平台距离水面1米，因此使用了两个3米长的PVC管道来操控装置在水中的移动，如上图所示。装置沿着平台侧面、面对乙烯基表面的方向水平移动。在PVC管道上设置了标记，作为视觉辅助工具，以保持装置在水平移动时保持相对恒定的深度。一旦到达乙烯基的末端，装置便直接垂直降至下一个标记的深度，然后沿平台水平移动，形成一种正方形的‘C’形状。数据采集从0.5米深度开始，返回至1米深度。我们提供的数据以ROS包的形式呈现，数据分为原始和预处理两种格式。原始数据未经编辑。预处理数据已同步，图像已进行立体校正和无畸变处理。与ROS包一起，我们还包含了立体相机和立体-惯性（相机-IMU）校准的yaml格式输出文件。相机和立体-惯性校准的原校准数据也包含在ROS包格式中，以便用户如需进行自己的校准。在姊妹GitHub的元数据文件夹中，我们以PDF形式详细描述了数据采集方法，以供用户对数据集进行背景理解，并提供使用的传感器数据表以及支持代码。