Estimation of A 3D Object Pose Using The Single Flexible Template

Recently, object recognition methods are widely developed in the field of industrial robotic vision since the task to take and place object from a picking to a placing station remains be a challenging topic. Various methods have been proposed, however having an inexpensive vision system by using single standard camera is another challenge. This paper presents an inexpensive vision method by using a standard monochrome camera to recognize and estimate a 3D pose object. The proposed algorithm utilizes a Directional Chamfer Matching technique where a template of database is generated from-a 3D CAD model, yielding a flexible usage. A technique named Outer Line method is also engaged to extract the contour of the object for the additional data of each template pose. The experiment both in using synthetic and actual image have been conducted, resulting less than 0.5mm of translation average error while for rotational average error is less than 3.5°. A further research will be studied with the aim allowing an industrial robot to select the best object from a box which contains similar objects.