Human Sensing Using Computer Vision for Personalized Smart Spaces

Smart spaces are everyday environments augmented with computing technologies that enhance human experience and activity performance. Continuous recognition of the presence of people, their identity, location, movement and activity patterns in real-time is a key challenge to address if smart spaces are to be envisioned as personalized and adaptive spaces. This paper introduces the multiple technologies available for human sensing and identification, discussing their advantages and disadvantages. In particular, Kitchen As-A-Pal is described as a smart space with real-time human sensing capabilities using computer vision by fusing fisher face recognition and skeletal tracking approaches. A wall-mounted Kinect is used for both single occupant and multi-occupant settings in kitchen As-A-Pal. The fused approach gives human identity recognition accuracy of 91.75% precision and 66% recall values for single occupant setting with good smart space coverage. Challenges do exist for human identity recognition in multi-occupant settings.