In this paper, techniques for tracking and grasping moving objects with an unknown speed on a conveyor using an eyein-hand robot arm are presented, which are useful in a production line for automatic object classification. First of all, the CAMshift (Continuously Adaptive Meanshift) algorithm is employed to continuously track a moving object in the image plane. Then, the minimum area rectangle method is integrated for correctly identifying a rectangle enclosing the target object. Object features for tracking purposes can be extracted from this rectangle. Next, through the application of an image Jacobian matrix, the tracking error in the image plane can be transformed to be the displacements of the robot’s end effector. Accordingly, the robot arm can be controlled for tracking this object. However, because of sensor noise and the fact that the object is moving, tracking errors cannot be eliminated at this stage. Therefore, the Kalman filter is used to estimate the state of the moving object, especially the moving speed. Finally, on the basis of the estimated speed, the robot gripper can thus be controlled to the point on the conveyor for accurately grasping and placing the moving object to a specified location. Experimental results showed the effectiveness of the techniques for grasping different target objects with different moving speeds and at any orientations.

Included in

Engineering Commons