Machine learning classification method for wheelchair detection using bag-of-visual-words technique.

Abstract

Purpose: The primary goal of this study is to enhance safety and accessibility for individuals using wheelchairs by enabling automatic wheelchair detection through a visual surveillance system. This contributes to the development of smart healthcare systems that facilitate autonomous navigation and improve mobility support.

Materials and Methods: A novel machine learning model based on the bag-of-visual-words (BoVWs) technique was developed for wheelchair detection. The approach involves key feature extraction, visual vocabulary construction, and histogram-based image representation. A support vector machine (SVM) classifier was employed to classify images based on these features after converting them into histograms of visual words. The model was evaluated using a publicly available image dataset.

Results and Conclusions: The proposed method achieved an accuracy of 98.85%, demonstrating its effectiveness in identifying wheelchairs in images. These findings highlight the potential of object detection techniques in recognizing mobility aids, contributing to improved accessibility and safety in rehabilitation and assistive technology applications.