The increasing mortality rate in motorcycle accidents has forced government to introduce helmets as mandatory safety equipment. In this work, an automatic helmet recognition system is proposed to group motorcyclists as helmet wearers and non-helmet wearers. The proposed system has three main steps, namely, moving object detection, motorcycle detection and non-helmet wearer detection. his work first extracts multiple features from the moving objects. The moving objects are extracted using frame difference method. During motorcycle and helmet identification, multiple features are extracted, from which optimal features are extracted using a method that combines Maximum Relevant Minimum Redundant algorithm with two hybrid methods combining genetic algorithm and ant colony optimization methods. The optimal feature vector is then used to train an ensemble system that uses SVM as base classifier. The performance of the ensemble system is improved through the use of a preprocessing step that used connected component labeling and visual features to first detect candidate motorcycle vehicles. Experimental results proved that the proposed algorithm improve both motorcycle and helmet detection and produced a high accuracy of 99.1% during motorcycle detection and 96.50% during helmet detection.

KEYWORDS : Automatic Helmet Detection, Automatic Motorcycle Identification, Multiple Features, Hybrid Feature Selection,Genetic Algorithm, Ant Colony Optimization, Ensemble Classification.