aUToLights Traffic Light Dataset
The perception system used to collect the aUToLights dataset. It contains a Novatel GPS/IMU unit and four cameras in total. The front-facing center long-range (second from left) and center wide-angle (third from left) are used to capture images of the traffic lights.
aUToLights is a traffic light dataset that can be used to train, validate, and evaluate traffic light detection and tracking performance on a multi-camera setup. The data was collected by members of aUToronto on the multi-modality perception system deployed on our new self-driving car, Artemis.
The primary sensors involved in collecting this dataset include a Novatel PwrPak7 GPS/IMU with TerraStar corrections (< 10 cm reported position error), two front-facing 7.1 MP Cameras (one with a 16mm f/1.8 lens as long-range, and another with a 8mm f/1.8 lens as wide-angle) from Lucid Vision Lab. The ground truth annotations are human labelled 2D bounding boxes with the aid of traffic lights’ state sequence from the light controller.
This dataset contains three types of traffic lights that conform with Michigan Department of Transportation, namely 3-light, 4-light of protected left, and 5-light doghouse with protected left.
Paper “aUToLights: A Robust Multi-Camera Traffic Light Detection and Tracking System” is under review.
To gain access to the dataset, please fill in the request form below.