Combination of HAAR, HOG, and LBP Descriptors for Enhanced Classification of Moving Objects and Motorcyclists Wearing Helmets
Abstrak
Many traffic accidents result in human casualties in many countries, especially for motorcyclists who do not wear protective helmets during the accident. In several regions, riding a motorcycle without a helmet is a legal offense. Reducing the number of helmetless motorcyclists can significantly save lives. Therefore, the development of an automated, real-time detection system can assist law enforcement in enforcing helmet use more effectively. Such a system demands high accuracy in identifying moving objects on highways and distinguishing between motorcyclists with and without helmets. This study employed feature extraction techniques, including HAAR, Histogram of Oriented Gradients (HOG), and Local Binary Patterns (LBP), which were concatenated to enhance classification performance. HAAR features capture contrast differences in rider images, HOG detects the shape of the rider’s head, and LBP analyzes the head texture. In addition, in this research, three classifiers were evaluated: Backpropagation Neural Network (BNN), Support Vector Machines (SVM), and Random Forest (RF). The results, based on a dataset of 1,956 images, demonstrated that the proposed concatenated descriptor achieved 99.75% accuracy in moving object classification using RF, and 96.75% accuracy in helmet detection using SVM and BNN. The method outperformed systems utilizing single or dual descriptors, indicating the effectiveness of the combined approach.
Identifier
- OpenAlex: https://openalex.org/W4411122985
- DOI: https://doi.org/10.48084/etasr.10677

