A robot detects its surroundings through camera information and its response requires a high-speed image process. Due to the increasing application of vision systems, various algorithms have been developed to increase speed of image processing. This paper proposes a double density Discrete Wavelet-based Neural Network to enhance feature extraction and classification of parts in each picture. The Discrete Wavelet-based Neural Network combines multi-scale analysis ability of the wavelet transform and the classification capability of the artificial neural network by setting the wavelet function as the transfer function of the neural network. The automatic assembly process needs to capture the image in an online process in order to recognize the parts in the image and identify the location and orientation of the parts. In this part, the two dimensional double density discrete wavelet transform have been applied to compress and remove noise from the captured Image. By applying a value for the threshold, the coefficients of the wavelet transform function are obtained using these coefficients and the characteristics of the wavelet coefficients are calculated. Subsequently, a multilayer perceptron is trained using these extracted features of the images. To find the best vector characteristics, various combinations of extracted properties have been investigated. This method has succeeded in object detection and results show that the Neural Networks and the training algorithm based on the wavelet transform function have exquisite accuracy in classification. Thus, the developed method is considered effective as compared to other state-of-the-art techniques. Manuscript profile

In this paper, an eye-in-hand stereo image-based visual serving controller for industrial 6 degrees of freedom manipulator robots is presented. The visual control algorithms mostly use the relationship between camera speed and changes in image features, to determine the end-effector movement path. One of the main problems of the classical IBVS method is the inability to estimate the distance of the object related to the camera, which requires peripheral equipment such as a laser rangefinder to estimate the depth. In this study, two cameras were mounted on the end-effector of a 6 DOF manipulator robot. The distance of the object to the camera is estimated by the equations associated with the epipolar plane, and the interaction matrix is updated at any time. For increasing response speed, the image interaction matrix was divided into two separate parts related to translational and rotational motion, and it was found that only the translational motion part is affected by distance. The control method separates the camera motion into three-stage based on pure rotation, pure translation, and hybrid motion, which has a better time response compared to the classical IBVS control methods. Additionally, a method for position prediction and trajectory estimation of the moving target in order to use in a real-time grasping task is proposed and developed using Recursive Least Square as the trajectory estimators in the image plane. The simulation results show that the proposed method increases the system response speed and improves the tracking performance. Manuscript profile