In the synthetic environment provided by Parallel Domain, traffic lights are accurately simulated to reflect their real-world counterparts, with vehicles and pedestrians behaving in a realistic manner when interacting with them. However, there is also the ability to customize certain scenarios, such as forcing vehicles to cross intersections at yellow or red lights, or programming pedestrians to [jaywalk](doc:jaywalking near intersections.
This flexibility allows for a wide range of training scenarios to be simulated. Additionally, Parallel Domain also includes HAWK lights, which are specialized traffic signals that help pedestrians safely cross the road.
There are a few use cases for synthetic traffic lights, some of which include:
Challenging weather conditions: Some customers are interested in synthetic data that simulates traffic signals in challenging weather conditions, such as fog or adverse lighting scenarios, to better train their AI systems to understand traffic signals under these conditions.
Partially occluded traffic signals: Customers may also request synthetic data that includes traffic signals that are partially occluded, to better train their AI systems to detect and understand these situations.
Uncommon traffic sequences: Customers may also request synthetic data that includes uncommon traffic sequences, such as sequences where all the bulbs on a traffic light are turned off for the duration of the sequence, to better prepare their AI systems to handle these unusual scenarios.
Our traffic light housings come in different shapes and sizes, and represent the vast majority of traffic lights found throughout the US and Europe. Below is a small sampling of our library.
Traffic light housings contain bulbs that match each traffic light type - for example, HAWK lights only contain yellow and red. In addition to the basic green/yellow/red bulbs, we also have special pedestrian walk signs, u-turns and bicycle signals.
We provide all our camera and lidar annotations for traffic light housings, including instance segmentations. In addition to providing bounding boxes around the traffic light housing, we also provide bounding boxes in 2D around individual bulbs. We also provide the bulb type: