A Robus Method for Wheelchair Detection: a Combination of the Gaussian Mixture Models and Histogram of Oriented Gradients

Abstract

An important function for a smart health care system, aiming to maximize safety and comfort to elderly or people with dysfunctional legs, is the automatic detection of a wheelchair captured from a visual surveillance system. In this paper, we proposed a method for detecting a two-dimensional wheelchair image using a combination of the Gaussian Mixture Models (GMMs) and the Histogram of Oriented Gradients (HOG). The proposed method consists of three main steps: (i). foreground segmentation, (ii). feature vector extraction, and (iii). wheelchair detection. The GMMs technique was used to extract a moving object from a background, while the underlying feature vectors of the moving objects were obtained using the HOG method. Finally, the Support Vector Machines (SVM) was implemented to classify a wheelchair object. We implemented 1,217 images for evaluating the performance of our proposed method which results in 86.01% of the accuracy rate. The advantage of our proposed approach is that it can detect a wheelchair effectively without any knowledge or prior information of the previous frames.