From Recognizing paintings wthout recognition (Russian) got idea about recognizing objects without neural networks but with OpenCV’s algorithms KAZE, SIFT, ORB. This might work at least for detecting road signs. But could work for recognizing as well.