Skip to content

Traffic Light Detection

Overview

trafficlight_1

  1. Obtain the rough ROI of the current traffic lights on the road through traffic light map based detector, camera intrinsics, camera extrinsics, and vector map.
  2. obtain the fine ROI through traffic light SSD fine detector based on rough ROI.
  3. classify the traffic lights using traffic light classifier based on the fine ROI.
  4. Return the traffic light classification results to behavior velocity planner in the Planning module to modify the trajectory speed.

How To Use

Draw Lanelet2 Map

You need to draw lanelet and their referenced traffic_light in the Lanelet2 map. Refer to lanelet2 map.

Calibrate Camera Intrinsics and Extrinsics

  1. For intrinsics calibration, refer to camera intrinsics calibration and save the parameters in ~/pix/pix-kit/Autoware/src/sensor_kit/external/pixkit_sensor_kit_launch/data/camera_info.yaml.
  2. For extrinsic calibration, refer to LiDAR-Camera extrinsics calibration and save the parameters in ~/pix/pix-kit/Autoware/src/param/autoware_individual_params/individual_params/config/default/pixkit_sensor_kit/sensors_calibration.yaml.

Set Goal Pose

Set the Goal Pose to generate a LaneletRoute. You can see a green line connecting the traffic light and the vehicle in the image below: trafficlight_detection_2

Traffic Light Detection

  • When a red signal is detected, the trajectory after the stop line becomes red, indicating that the vehicle will stop before the stop line trafficlight_detection_3
  • When a green signal is detected, the trajectory after the stop line becomes yellow or green, indicating that the vehicle will pass the intersection with traffic lights. trafficlight_detection_4

Reference