Fast and reliable two-wheeler detection algorithm for blind spot detection systems

Abstract

In this paper, we propose a real-time detection algorithm using a MCT AdaBoost classifier which detects two-wheeler in a blind spot. The proposed algorithm uses a cascade classifier generated by AdaBoost learning based on the MCT feature vector. The MCT AdaBoost classifier is composed of weak classifiers as many as the number of pixels of the detection window, and each pixel becomes a weak classifier. The smaller the detection window, the faster the processing speed, and the larger the detection window, the greater the accuracy. The proposed algorithm uses two classifiers with different detection window sizes. The first classifier generates candidates quickly with a small detection window. The second classifier verifies the generated candidates with a large detection window. Accordingly, the proposed algorithm supports fast and reliable two-wheeler detection. Also, the proposed algorithm uses a wheel classifier in order to detect an adjacent two-wheeler in the blind spot which is well not detected by two-wheeler classifiers. Experimental results show that the proposed algorithm has faster processing speed and higher detection rate than a single classifier without generating candidates.