抄録
We examined the detection method for wake-up behavior using visible ray camera to predict patient's ambulation behavior in advance. In our past studies, we proposed a method that configured mask patterns by using the edge instead of CHLAC feature. The study furthermore showed that the wake-up behavior was detected with high accuracy compared to using CHLAC feature by combining with the process where the abrupt slope changes were dealt as noise in the mask pattern configuration. However our conventional approach did not deal the cases where the edges with different gradient directions on the same object presented the same movement. For example, when a boll is moving, each outer edge has different gradient direction but the moving direction is the same. Since each of the movements of the outer edge is the information that expresses the movement of the ball. itself. In this paper, we propose the method where, by convolving the feature vectors of the same movement without performing the conventional noise reduction process, they are dealt as the same feature. The examination result has confirmed the detection rate of our method has improved by 25[%] compared to the conventional method.
