A Robust Object Recognition Method Using Surface Images of 3D Objects

Robots which automatically perform outdoor work re-quire robust and accurate environments. The authors de-vised a new object recognition method, comparing surface images of a target object with 2D surface image models stored in a database. First, range data is segmented into several regions to obtain stable surface images. Second, a surface image is compiled from each stable region. The surface image is comprised of a series of depth values. The depth value is defined as the height from a tangential plane on the center point of a region to points on a geodesic cir-cle. Third, the compiled surface image is compared with prototypical surface images, so as to identify the classifica-tion of the region. Lastly, the object’s pose can be precisely obtained by applying the 3D template matching technique with the use of a 3D object mesh model. From these pro-cesses, the proposed method enables robust recognition of a complex-shaped object. Through the simulation applying the synthesized 3D model, as well as the outdoor experi-ment, the validity of the proposed method can be obtained. The method is expected to allow robot operators to perform distribution work, such as pole and insulator replacement, more easily and rapidly.