| 摘要 |
<p>According to the present invention, a method for segmenting and recognizing an RGB-D image-based object for 3D objection recognition comprises steps of: calculating a normal vector for each pixel, based on a 3D image including depth information acquired through a camera; clustering the normal vector calculated by using a K-means clustering algorithm to the preset number of planes; calculating a plane normal vector respect to an area on which user interrupt occurs in a 3D image in which the normal vector for each pixel is clustered to at least one or more number of planes; dividing a reference plane including the area on which the user interrupt occurs from a background screen through user interaction, based on the calculated plane normal vector; and constructing a convex hull with respect to the reference plane and calculating number and position of an object with respect to contours which are extracted based on specific point clouds data.</p> |
