SAIT Traffic Landmark Dataset

We disclose a traffic landmark dataset for detection.The dataset generated with our framework includes about 150,000 images and annotations of about 470,000 traffic landmarks.Our dataset was collected in an urban area of Seoul and suburban areas of Suwon, Hwaseong, Yongin, and Seongnam in South Korea at different times of the day.Images taken in the morning or evening included a large number of saturated areas due to exposure to direct sunlight.Most images taken under the light condition of the late evening was low-contrast.The images taken at noon included the reflection of the windshield due to strong sunlight.These various elements contribute to the development of a robust detection algorithm because they can be encountered in a real test environment.

本团队公布了一项针对交通标志检测的交通地标数据集。该数据集采用我们的框架生成，包含约150,000张图像及约470,000个交通标志的标注。数据集收集于韩国首尔市区及水原、华城、龙仁、城南等近郊地区，采集时间跨越白昼的不同时段。清晨或黄昏时段拍摄的图像因直射阳光照射，大量区域呈现饱和状态。晚间较晚时刻拍摄的照片由于光线条件，对比度较低。正午拍摄的图像则因强烈阳光，出现了挡风玻璃的反光。这些多样的因素有助于构建一个稳健的检测算法，因为这些元素在真实测试环境中均有可能出现。