An augmented reality device based on recognition of a spatial structure includes: a point cloud normal vector extracting unit extracting a normal vector for a point cloud from image data input from a camera; a plane object segmenting unit segmenting a plane in the image data by using the extracted normal vector; a representative plane selecting unit selecting a representative plane among the segmented planes; a spatial structure extracting unit recognizing a spatial structure by using the representative plane; and a virtual object matching unit matching a virtual object in the recognized spatial structure.