Traffic Light Dataset in SOUTH Korea
Traffic Light (TL) in South Korea
In contrast to Europe and the USA, most TLs for vehicles in South Korea at intersections have a horizontal layout and are installed as side-pillar horizontal types. A TL can have three or four signals and one signal consists of a 355 mm x 355 mm black box with colored bulbs. The diameter of the bulb is 300 mm. There are two types of bulbs: a circle and an arrow. The circle bulb indicates green, red, and yellow, whereas the arrow bulb represents a left turn. There are two combinations for the three bulb TL, and there is one type for the four bulb TL. The TL status can be green, yellow, red, green + left turn, and red + left turn.
Traffic Light Dataset in South Korea (TL-Seoul)
A traffic light recognition is one of the challenging parts for autonomous driving. Unfortunately, there is no public dataset for South Korea. In addition, the type of TL in South Korea is horizontal as opposed to the TLs in Europe and the US. Furthermore, road conditions in South Korea might be one of the toughest environments to recognize TLs due to the background complexity and great amount of clutter in dense urban settings. To stimulate relevant research activities in our country and bring the attention of many researchers from other countries, we built our TL dataset of Seoul in South Korea. Seoul is the capital city, with a 0.6 percent area of South Korea and 20 % people (9.7 million) residents. Seoul consists of a total of 25 districts, and each district has an independent district office. The dataset consists of images acquired along a route between the 25 district offices.
The camera used to acquire the data was an acA2000-165uc from Basler and was recorded at 30 fps with a 2040x1086 resolution and a 16 mm lens. It is composed of 483 sequences and the images are stored as a dataset for every ten frames. The left figure shows the position distributions of all the TLs in the dataset and the frequency of the TLs according to the height of TLs is drawn in the right figure.
The total number of frames is 25,882 containing 75,255 TLs, and the dataset is divided into three parts for training, validation, and testing as in the below table. The dataset is available at this download link (20 GB).