Fall Detection With Video Surveillance Systems

Facing with the growing population of seniors, the society needs to pay more attention to thelife quality and health condition of the seniors. What s more, falls in the elderly population have ahigh incidence and a more serious consequence. Given the the great damage of the seniors s bodycaused by the fall, it is urgent to develop a efficient fall detection algorithm, which can effectivelyreduce the harm to the old people’s physical and mental health and the corresponding economicpressure.At present, the existing majority of fall detection system are based on sensors.Their havehigh prices and require to wear a particular device which can bring discomfort to theelderly.Compared with the fall detection system based on sensors, video surveillance system has agood user experience and lower cost which can take advantages of parallel tasks.In this thesis, we propose two fall detection algorithms based on video surveillance system.Oneof them is an adaptive fall detection approach based on inner distance shape context method inhome environment. This method is based on analyzing human shape deformation during a videosequence.A inner distance shape context method is used to track the person s silhouette along thevideo sequence. The shape deformation is then quantified from these silhouettes based on shapeanalysis methods. Finally, falls are detected from normal activities using dynamic time warpingmethods. The other fall detection algorithm is based on human shape analysis.Through the analysisof the body shape, the algorithm extract the global motion of human body, human head movementinformation, the projection histogram features. With all these features the algorithm judges a fall bya machine learning method.Experiments show that the first approach has a lower time complexityand can quickly determine fall; the second method has better adaptability to complex condition andhas higher precision compared with the method based on threshold value.