Workpiece Recognition Technology Based on Improved SIFT Algorithm

Abstract

AbstractTo solve the problem of workpiece image matching and recognition in complex environments such as rotation, scaling, and partial occlusion, and to solve the problems of slow running speed and low matching accuracy of SIFT algorithm, a scheme of reducing the dimension of feature point descriptor and selecting the best feature point matching is proposed. A concentric ring with a radius of 2 pixels, 4 pixels, and 6 pixels is constructed with the feature point as the center. According to the rule, the pixels close to the feature points significantly impact the features. The pixels far from the feature points have a minor impact on the features. They are divided into 4, 8, and 16 partitions, respectively, generating 56-dimensional descriptors. The feature points are divided into two sets in image matching: maximum and minimum. The distance between the feature points of the same typeset between the two images is calculated to select the best matching points, which reduces the amount of image matching calculation and saves the algorithm time consuming. The experimental results of workpiece image recognition show that the improved algorithm of feature point descriptor dimensionality reduction and image matching can effectively improve the accuracy and reduce the image matching time.KeywordsSIFT algorithmFeature point descriptorFeature matchingImage matchingWorkpiece recognition