The simplest ADAS that detects and classifies road signs from the camera's video stream in real-time.

Cascade classifiers (two classifiers: for circular and for triangular and rectangular traffic signs) trained on LBP-features is used to detect road signs. The detected road signs are then filtered by the linear binary SVM-classifier (also two classifiers: for circular and for triangular and rectangular traffic signs) trained on HOG-features. Finally, the common multiclass SVM-classifier (one-vs-all scheme, with RBF kernel) learned on HOG-features predict classes of detected traffic signs.

Input image: size=1024x768
Pyramid of images: scaleFactor=1.2
NMS: minNeighbors=5
Sliding window: minSize=24x24, maxSize=144x144

Cascade classifier for circular traffic signs:
recall=76.8%, precision=78.5%, f1-score=77.6%
Cascade classifier for triangular and rectangular traffic signs:
recall=80.0%, precision=81.6%, f1-score=80.8%
SVM-classifier for circular traffic signs:
accuracy=98.6%
SVM-classifier for triangular and rectangular traffic signs:
accuracy=97.46%

SVM-classifier:
accuracy=96.5%

CPU: Intel Core i5-4570
RAM: 8Gb
Inference: 9-11 FPS (single thread)