工件图像匹配的局部敏感哈希应用改进

An Improvement on Workpiece Image Matching via Locality Sensitive Hashing

摘要: 工件轮廓的提取与匹配是机器人分拣中最重要的一环。传统的轮廓匹配是采用Hausdorff距离进行模板匹配，但是计算量过大且要求工件的位置和角度一定。本文在模板匹配的基础上，采用图像特征向量进行图像匹配，降低了计算量并解决了工件的几何变换带来的图像识别问题。改进传统的线性搜索，采用敏感哈希算法匹配图像，并且对哈希算法做出了一定的改进，降低了存储空间。通过实验发现，本文所采取的算法可以有效的忽略在图像识别中工件平移、缩放、旋转所带来的问题，并且在一定程度上提高了图像搜索的效率和准确度。

Abstract: The extraction and matching of workpiece contour is one of the most important steps in robot sorting. The traditional contour matching algorithm uses Hausdorff distance, with considerable amount of calculation and difficulty in determining the position and angle of the workpiece. Based on the template matching, the image feature vector is used to match the image in this paper, which reduces the computation and solves the problem of image recognition. We also improve the traditional linear search algorithm by using the Locality Sensitive Hashing algorithm. Moreover, the hash algorithm is improved to reduce the storage space. In the experiment, the algorithm can effectively ignore the problems caused by the translation, zoom and rotation of workpiece during image recognition. What’s more, it can improve the efficiency and accuracy of image search to a certain extent.