Closed fscbr closed 7 years ago
I added a finder method to search for possible traffic light positions. In the simulator that works well. The images can be reduced to 128*128 at least and the created ground truth imgaes seem to be well centered
I committed 64x64 bit ground truth images and a trained SVM. Accurace is 99% so nearly perfect but for simulator only
currently the ground truth images are 256*256 large and contain a not centered traffic light. the existing svc classifier could be used to find the center position of the traffic light which could be the cropped to 64x64. That would help to improve the accuracy of the traffic light classifier