Visual based fall detection through human shape variation and head detection

In this paper, an improved method is proposed to detect falls using an uncalibrated camera. The proposed fall detection technique combines human shape analysis and human head detection together to detect falls from normal daily activities. The human shape is represented with an ellipse shape and features extracted from the ellipse are used to detect fall events. The head detection helps to distinguish between falls and fall-like incidents in the case where the activities of daily living at home happen to be parallel to the camera optical axis. Two novel approximate human head shape models are proposed to detect the head of the person. The experiment results demonstrate that this proposed method is able to achieve high detection accuracy compared to other methods in the literature.