Traffic Light Detector
AutonomouStuff’s Traffic Light Detector software leverages the power of deep neural networks (DNN) to enhance open-source autonomy stacks with the ability to detect and classify traffic lights. It was built within the framework of a Robot Operating System (ROS) and Autoware to add critical functionality across a broad spectrum of driving environments — from sparse, single-signal intersections to dense urban corridors. The software together with Autoware works to detect the presence and location of the traffic light nearest to the lane in which the vehicle is traveling, then isolates that signal in a cropped image to determine the state of that particular traffic light.
Key features of the software include:
- expanded automated vehicle functionality.
- compatibility with Autoware and Robot Operating Systems (ROS).
- ability to navigate sparse, single-signal intersections and dense urban corridors.
- detection of presence and location for nearest traffic light to lane of travel.
- isolation of nearest signal for traffic light state classification.