Xanadom1/auto_drive_traffice_light

Buckeye AutoDrive Traffic-Light States

Images of traffic lights, each labeled with the light's state. I built this dataset on the perception team of Buckeye AutoDrive, The Ohio State University's team in the SAE/GM AutoDrive Challenge II, advised by Prof. Wei-Lun Chao.

Contents

• 9,930 images: 7,943 for training and 1,987 for validation.
• Three classes:
  • Black_None: [the light is off / no lit light is visible].
  • Green Light: a green light.
  • Red Light: a red light. [It also includes yellow lights. We merged red and yellow into one class because both mean the car must brake.]
• Format: ImageFolder, one image and one label per example.

How it was made

The images come from video recorded on our team's car[ on ___]. I labeled the video frame by frame[ and cropped each traffic light from the frame].

Model

A YOLOv8n classifier trained on this data: Xanadom1/AutoDrive_TrafficLight_Baseline_V1

Author

Fengze Jia, The Ohio State University · fengzejia.com