Fall Detection with Part-Based Approach for Indoor Environment

In the current scenario, majority of the aged people want to lead independent life, and most of them prefer living at their own home. According to recent case studies, the major cause of casualty among elder people has been due to the accidental falls. Hence, it is eminent to have a fall detection monitoring system at home. The prevailing method for fall detection uses accelerometers to distinguish fall from other day to day activities, these results are more erroneous. In this paper, vision based "Fall detection with part-based approach (FDP)" is proposed to give accurate information about the person activities in the indoor. The proposed scheme uses background subtraction in association with aspect ratio and inclination angle to detect the fall. Moreover, the proposed approach predicts the fall even if the person is occluded by other objects or under self-occluded condition. To detect the person even if only partly visible and occluded by other non-moving objects, part based approach is adapted. To train the system for detection purpose, Cascaded structure of Haar-rectangular features with joint-boosting classifier is utilized. The detection efficiency is measured by precision, recall and accuracy parameters.