An edge feature-based approach for workpiece localization and determination of feasible clamping regions

We present an approach for automatic grasping based on object recognition and localization using the maximum edge of workpiece. This approach firstly extracts the sum coefficient matrix in the horizontal, vertical, and diagonal by the first layer of high frequency of wavelet decomposition structure. Find the key points that could satisfy the requirements of being the local maximum value and greater than the given threshold. Searching the points located in the maximal edge of workpiece from the set composed of key points by polar coordinates transform. By linear interpolation method, obtain the radius under the radians defined beforehand. Artifacts teach an initial position and angle, and calculate the maximal edge. Matching each maximal edge of new workpiece with the maximal edge of initial workpiece, calculate the rotation angle current workpiece relative to the initial workpiece. Aiming at the classification of the workpiece, Principal Component Analysis (PCA) algorithm can be applied to compute the eigenvalues which are the rotational invariant. The vector being composed of main eigenvalues can be used to solve the problem of classification.