The method is presented for object assembling via a manipulator robot. This method was designed to track the pieces based on image feedback for pick and placement tasks. The depth of the pieces is calculated by stereo triangulation. Vision-guided robotics is most often based on the training steps, therefore it is a time-consuming process. For this reason, we have proposed a modified stereo vision method to predict the movement of the part in the image space. The linear and angular velocities of moving objects predict by a state estimator. The object velocity components predicted by estimation algorithms such as Kalman filter, Recursive least square and Extended Kalman Filter. Results show that in the case of, Extended Kalman filter estimator shows better tracking and convergence behavior. This method does not need to know the three-dimensional model of the parts, and it can be used on pick and place robots if the physical limitations of the joints are considered. The proposed method was experimentally tested in a laboratory environment. The cameras were installed parallel and non-parallel to determine the effect of the field of view on the precision and speed detection. A comparison of the simulation and experimental results showed that the use of the parallel stereo Image-based visual servoing with the Extended Kalman Filter method could be smoother and more accurate than the other methods.