车辆检测数据集ITCVD(Vehicle)

TCVD数据集是一个大规模、注释良好且具有挑战性的车辆检测数据集。
这些照片是从荷兰恩斯赫德上空飞行的飞机平台上拍摄的，高度约为330米。这些照片采用了底视和斜视两种视角。
科视的倾斜角度为45度。底部图像的地面采样距离为10厘米，
总共选取了173幅图像，其中135幅图像(23543辆)用于培训，其余38幅图像(5545辆)用于测试。
数据集中的每个车辆都使用限界框进行手动注释。
地面真实数据的矩阵大小为Num vehiclexNuminformation，其中Num_vehicle表示当前图像中的车辆数量Num information表示每个标记车辆的信息。在这个数据集上，Numinformation的长度为6。
第一列和第二列表示限界框(x，y)左上角的坐标。
第三和第四列描述了限界框(x，y)右下角的坐标。
然后自动计算限界框中心点的坐标。
根据车辆的中心和一个附加点来标记车辆的方位。
比附加点作为车辆方位信息存储在数据集中的第五和第六列中。

The TCVD dataset is a large-scale, well-annotated, and challenging vehicle detection dataset.
These photos were captured from an aircraft platform flying over Enschede, the Netherlands, at an altitude of approximately 330 meters, adopting two perspectives: nadir and oblique. The oblique viewing angle is 45 degrees. The ground sampling distance (GSD) of the nadir images is 10 centimeters.
A total of 173 images were selected, among which 135 images (containing 23,543 vehicles) were used for training, and the remaining 38 images (containing 5,545 vehicles) were used for testing.
Every vehicle in the dataset is manually annotated with bounding boxes.
The ground truth data is stored in a matrix with dimensions of Num_vehicle × Num_information, where Num_vehicle represents the number of vehicles in the current image, and Num_information represents the information for each annotated vehicle. For this dataset, the length of Num_information is 6.
The first and second columns represent the (x, y) coordinates of the top-left corner of the bounding box. The third and fourth columns describe the (x, y) coordinates of the bottom-right corner of the bounding box. The coordinates of the center point of the bounding box are then automatically calculated. The orientation of the vehicle is annotated based on the vehicle's center and an additional reference point. This additional point is stored as the vehicle orientation information in the fifth and sixth columns of the dataset.

搜集汇总

背景与挑战

背景概述

ITCVD是一个大规模车辆检测数据集，包含从飞机平台拍摄的173张航空图像（135张训练，38张测试），覆盖荷兰Enschede地区，拍摄高度约330米，提供底视和斜视（45度倾斜）视角，地面采样距离为10厘米。每辆车均使用边界框手动标注，并包含车辆方向信息，适用于车辆检测算法的训练和评估。