A FLEXIBLE RELATIONAL FEATURE MODEL FOR FALL DETECTION

摘要

Vision based fall detection solutions support elderly whornlive alone in their homes. For people falling down and notrnbeing able to stand up on their own again, such a fall isrna major risk. In this work, we show a person detectionrnapproach using a relational feature model using consumerrndepth cameras. We propose a flexible relational featurernmodel (FRFM) for fall detection in combination with histogramrnsimilarity functions such as the bhattacharyya distance,rnhistogram intersection, histogram correlation and thernchi-square χu001f2 histogram similarity function. FRFM is anrnextension of the relational feature model (RFM) with thernadvantage that it can be used for all rotations of a body. Thernextension is necessary for fall detection due to the fact, thatrna lying person is rotated in the image where the standardrnperson detection approach detects upright standing personsrnonly. The relational features are computed for verificationrnsituations during fall detection on the basis of the Histogramsrnof Oriented Gradients (HOG) feature descriptor.rnThe experimental results show the best setup parametersrnfor our feature model with different types of images (RGBrnand depth) and results on the specific field of fall detection.