Evaluation of a video based fall recognition system for elders using voxel space

—Video is a rich source of information that can be used to passively monitor the activity of elders. The amount of information contained in video is significantly greater than other sensing technologies such as RFID tags and motion sensors. Privacy of residents is preserved by not using the raw video, but instead, extracting binary silhouette maps, which represent the pixels a person occupies in an image. Silhouettes acquired from multiple cameras viewing the same scene are used to build a three-dimensional object whose activity is linguistically summarized for activity monitoring. These linguistic summarizations are used for abnormal event detection, specifically for the automated detection of falls. In this paper, we present three measures for system performance evaluation and discuss successes and difficulties in video-based human activity recognition of falls.