Objective. Elderly people, who occupy nursing homes with dementia or legs getting weak, are on the increase. The nursing homes are requested to keep wandering, fall or other incidents of residents from going to fatal accidents. We propose a new hazardous behavior detector with face recognition technology in nursing home corridors.
Main content. This system consists of some multi-camera gates to acquire corridor passer events and a processor for face recognition and behavior detection or estimation. Each camera gate is composed of network cameras, including front cameras, side cameras, and top cameras. We place these gates to corridor areas outside private rooms in the nursing home. The faces of the passing residents, care staffs or visitors are detected and identified with the front cameras or side cameras. The face recognition leads to the acquisition of 3-D passing position, passing speed, and passing time of the residents. From these transfer information, we assemble a state transition diagram of each resident for the gates and estimate who is in danger of wandering indoors or going out of the facilities without notice. We also detect the tumbles or the crouches around each gate with if-then rules from passing information.
Results. We assembled a prototype of the multi-camera gate equipment and conducted on-the-spot experiments at a nursing home in Japan. Eight incidents of wandering at daytime or late night that care staffs recorded, were all successfully located over 582 hour period and came down to a care staff post using a computer network. In the time period, 13 incidents of wandering whose documents did not exist, were also located. This out-of-record number leads work load reduction of care staffs.
Conclusion. We have proposed a hazardous behavior detector of residents with multi-camera gates in nursing home corridors and confirmed its performance with on-the-spot experiments. This system is useful when the care staffs are not abundant at late night or they are in shifts.