Detection of helmets on motorcyclists

The use of motorcycle accidents has rapidly increased. Although the helmet is the main safety equipment of motorcyclists, many drivers do not use it. This paper proposed a method for motorcycle detection and classification and a system for the detection of motorcyclists without helmets. For vehicle classification, we have employed the wavelet transform (WT) as the descriptor and the random forest as the classifier. For helmet detection, the circular Hough transform (CHT) and the histogram of oriented gradients (HOG) descriptor were applied to extract the image attributes, and the multilayer perceptron (MLP) classifier was used to classify the objects. The results for vehicle classification achieved an accuracy rate of 97.78 %. The algorithm step in the helmet detection accomplished an accuracy rate of 91.37 %. The results were obtained with the author’s database.