Human Posture Recognition
We develop lightweight algorithms that can reliably identify a person's posture in a visual scene. The algorithms are based on synthetic models of the mammalian visual system. See the syntethic vision section of our website for details on our work on bio-inspired object categorization. This algorithm can be implemented in embedded platforms, such as sensor networks and cellular phones.
Here is a link to our demonstration of the ultra-low computation posture recognition engine on a Apple iPhone cellular phone platform:
This video shows how the algorithm reliably recognizes posture for extended periods of time. Here we recorded a behavioral sequence of human postures for a 25+ minutes interval. A plot of the time sequence of postures from this video is given in the figure below.
Older Videos and results
Here is a link to a video demonstrating our posture recognition engine:
This demonstration shows our bio-inspired object recognition code performing rel-time identification of a person's posture on a laptop computer. This is targeting an assisted living application for home care and monitoring of elderly people or patients. The algorithm is described in a recent BioCAS paper.
This algorithm is lightweight and can be implemented in embedded platforms, as sensor networks and cellular phones. Here is a link to our demonstration on a Nokia Symbian S60 cellular phone platform of the ultra-low computation object/posture recognition engine: