Human fall classification system for ceiling-mounted kinect depth images
Sowmya Kasturi, Kang-Hyun Jo · 2017
The number of elderly people living alone has been considerably increased over the past few years. Hence the research regarding Ambient Assisted Living (AAL) systems has been given significant importance to improve the quality of life for them. Falls have become one of the major health concern among elders. Many fall detection and classifications methods are being developed to provide a reliable solution. The proposed system presents a vision based human fall classification method to discriminate falls from non-fall events. The depth images from a ceiling mounted Kinect camera are considered in the proposed system to preserve privacy, reduce the influence of occlusion and complex cluttered background. Human silhouettes are obtained after background subtraction and shape based features are extracted. A binary Support Vector Machine(SVM) classifier fed with these features is used to classify the fall events from non-fall events. The proposed method was tested on a publicly available dataset and classifies falls from other actions with an accuracy of 93.04%.