METHOD OF HUMAN ACTION PATTERN RECOGNITION USING HISTOGRAM OF MOTION INTENSITY AND DIRECTION FROM MULTI VIEW

摘要

According to an aspect of the present invention, provided is a method for recognizing an action pattern by using a histogram for intensity and direction of action from a multipoint of view. The method includes: (A) a step of collecting multi-view point images of an actor from multiple cameras arranged at angles distributed in the whole space which should be recognized in different directions; (B) a space normalization step of performing binarization with respect to each image input from the multipoint of view and performing a cropping process for trimming each image into a certain size focusing on the actor; (C) a time normalization step of resampling a sequence at the same interval and in the same frame with resect to the spatially normalized image; (D) a motion flow formation step for forming a motion flow having each center frame and adjacent frames after the time normalization step; (E) a step of dividing the image into lower level areas by using a grid; (F) diving the lower level areas into bin areas; (G) an action direction histogram generation step of generating an action direction histogram by calculating a predominant direction of the action from directional histogram bins divided in the lower level areas in step (E); and (H) an action intensity histogram generation step of calculating a ratio of the sum of intensity in all directions with respect to the sum of intensity in the predominant direction of action and converting the calculated ratio into a histogram.