Fall detection method based on semi-contour distances
Shanshan Xu, Xi Cai · 2018
In recent decades, as aging population and empty-nest families are increasing, falls in the elderly have become the health care issues that cannot be neglected. The computer vision-based methods are promising for fall detection, for they apply a camera installed indoors to monitor the movement of the elderly in real time. However, the features used in previous methods lack specificity, thus they have difficulty in distinguishing falls from some normal activities, such as crouching down. In this paper, we propose a novel fall detection method based on semi-contour distances for monitoring elderly people in the home environment. Firstly, we use the Gaussian Mixed Model (GMM) method to extract the human silhouette; then capture the geometric features, i.e. the semi-contour distances of the human silhouettes; finally apply the SVM classifiers to classify different actions to complete fall detection. The experimental results demonstrate that the accuracy of this algorithm is 86.79% and the sensitivity is 96.87%. It is shown that the proposed video-based fall detection method could achieve a reliable result.