Paper Title
Traffic Light Detection And Recognition Using Adaboost Classifier Based On Haar-Like Feature

Abstract
This paper proposes a method for detecting and recognizing traffic lights using a Haar-like feature based on an AdaBoost-trained classifier. The method of detecting traffic lights using only the existing classifier in an image is not suitable for real-time processing when a vehicle is in motion. In this paper, we set the ROI as the whole image to narrow the search area for the traffic light and to extract candidates based on the HSV color model in order to reduce the classifier�s detection and recognition time for traffic lights. In addition, the traffic light is detected and recognized accurately in the candidate region using a Haar-like feature�based AdaBoost classifier, which extracts the edge characteristics of the traffic lights based on the brightness difference in the estimated area. The proposed method significantly improves the False Positive detection rate (classifying objects that are not traffic lights as traffic lights) for driving images in an urban setting. The real-time system showed a detection rate of 98% and an average processing speed of 50 fps. Keywords- ADAS, Traffic Light, AWB, Machine Learning, Haar-like feature