Monocular camera based object recognition and 3D-localization for robotic grasping

Visual Servoing for a robotic hand is still a difficult problem to solve and is the topic of current research. To recognize an object of interest and calculating its orientation and location without extensive training is another far-fetched problem faced by researchers in this field. In this paper, object recognition based on scale-invariant feature based transform (SIFT) is done, which is used to calculate the location in image plane, scaling factor and orientation of the object in the testing image with respect to the training image of the object. Here the system is trained using single image instead of many images at different orientations. Further, the object location from 2D image coordinated is mapped to real world 3D coordinates where the third dimension is calculated using relative scaling obtained from SIF T. The i mages are obtained u sing a stationary monocular camera system. For this, we have used NAO which is autonomous, programmable humanoid robot developed by Aldebaran Robotics, France.