This thesis presents a monitoring system of falls for homecare and ward care applications. This system integrated with alarming devices and SMS (or MMS) services is able to mitigate the burden of caregivers and avoid the possible loss of life caused by delaying the treatment. The functions of this system is firstly to analyze the pictures divided from the video catched by webcam, then apply background subtraction method and binary threshold filter for protecting background constructions. As the image subtracting processes will generate noise, therefore, median filter is applied for noise elimination. After that, image projection filter is utilized to filter the size and shape of the object and its boundary coordinates for detection in human-shape. Finally, we will be able to judge the human behavior of falls, for example, by detection in human-shape and immediately trigger the alarm system. The key point of the falls monitoring system is to find the numbers of non-zero pixel value as reference for filtering the unwanted objects. The pixel value is tunable following the changing environments. The smaller the value, the higher the sensitivity. This way of detection is able to filter out the desired objects and make a judgment on unusual behavior quickly. Simulations of different human behaviors have been done include standing, prostration, squatting and falling down with different angles. Our experimental results show no error of judgement and conform that this monitoring system of falls is effective for home and ward care for those people who are in need.
