tl_detector

Traffic Light Detection

This node is in charge of detecting the traffic lights from the images fed from the on-board camera, when a red light is detected a message is published on the /traffic_waypoint topic.

The approach taken for this task is to train a neural network for object detection using each light state as a different class, in this way the classification task is reduced into the same network.

The documentation and code used for training can be found here.

The data for training was collected from both the simulator, recording a bag and then saving the camera images and from a bag provided by udacity that can be downloaded from here.

Additional bags are provided by udacity for testing the detector in the real world that were recorded at the Test Site.

To record a bag while running in the simulator:

$ mkdir ~/bagfiles
$ cd ~/bagfiles
$ rosbag record /image_color

After the bag was recorded we can play it running roscore in one terminal:

$ roscore

In a separate terminal we play the bag:

$ rosbag play -l bagfile.bag

and finally in yet a separate terminal we can save the images received from the /image_color topic:

$ rosrun image_view image_saver _sec_per_frame:=0.01 image:=/image_color

NOTE: The topic from the the ros bag recorded from the Carla at Udacity is instead /image_raw