BT-Crack2000 pavement crack semantic segmentation dataset

This study constructs a crack image dataset based on urban roads in Baotou City, China. The data collection was conducted using an onboard camera for field shooting, acquiring a total of 2102 high-definition road images. To enhance data diversity, the original dataset was expanded to 6500 multi-resolution images through data augmentation techniques such as image translation, selective cropping, and scaling. This dataset exhibits the following characteristics: (1) It includes a variety of background interference factors, such as lane markings, lighting variations (dim and glare conditions), and shadows in complex scenes; (2) These interference factors significantly increase the difficulty of image segmentation tasks, providing a challenging test benchmark for evaluating algorithm robustness. The construction of the dataset fully considers various interference factors in real road environments, offering reliable experimental data support for road crack detection research.

本研究构建了一套基于中国包头市城市道路的裂缝图像数据集（crack image dataset）。数据采集采用车载相机开展实地拍摄，总计获取2102张高清道路图像。为提升数据集多样性，本研究通过图像平移、选择性裁剪与缩放等数据增强（data augmentation）技术，将原始数据集扩充至6500张多分辨率图像。该数据集具备如下特征：(1) 包含多种背景干扰因素，例如车道标线、光照变化（昏暗与眩光场景）以及复杂场景下的阴影；(2) 此类干扰因素显著提升了图像分割（image segmentation）任务的难度，可为评估算法鲁棒性（algorithm robustness）提供极具挑战性的测试基准。本数据集的构建充分考量了真实道路环境中的各类干扰因素，可为道路裂缝检测相关研究提供可靠的实验数据支撑。

搜集汇总

背景与挑战

背景概述

该数据集是一个用于道路裂缝语义分割的挑战性数据集，基于中国包头市城市道路构建，包含6500张多分辨率图像，通过数据增强从原始2102张图像扩展而来。其特点在于包含了车道标线、光照变化和阴影等多种真实环境干扰因素，显著提升了分割任务的难度，旨在为算法鲁棒性评估和道路裂缝检测研究提供可靠的实验数据支持。