create training data TL Detection

[fscbr]

let the car drive around and record train images required to train the classifier.

[fscbr]

most work is done in the "create TL ground thruth" task. The detector is able to create sub images using the config list of traffic light way points. The more far the car drives, the more data will be available.

The next step is to save the images as its done in the "create TL ground thruth" task in a train data folder.

[fscbr]

tl_detector is able to produce data and save it. heading of the car must be close to the waypoint orientation, othervise the image does not contain the traffic light. If this is a problem, more work has to be spent to find the angle where to look to the left or right