摘要
针对传统零件识别系统时间效率低等问题,提出基于SURF(Speeded-Up Robust Features)特征的零件识别算法。该算法首先对零件图像进行相关预处理,然后通过SURF算法得到零件图像的特征点和相应的特征向量,最后根据零件识别任务的特点,运用近似最近邻算法ANN(Approximate Nearest Neighbor)进行特征向量的匹配搜索,得到与模板图像最相似的图像序列。通过SURF算法和ANN算法的联合使用,解决了零件在旋转、尺度、模糊和光照变化后的识别难题,实现了算法的实时性和工业化要求。实验表明,相比于传统零件识别算法,该算法在算法效率及稳定性等方面都有显著性提高。
Aiming at low time efficiency problem of traditional component recognition system, we propose the component recognition algorithm which is based on speeded-up robust features (SURF) technique. First, the algorithm makes correlated pre-proeessing on the component image, then it extracts the feature points of the component image and obtains corresponding eigenvectors by SURF algorithm. Finally, according to the features of component recognition task, it runs the approximate nearest neighbour (ANN) algorithm for matching and searching of the eigenvector to obtain the image sequence mostly similar to template image. Through the joint use of SURF and ANN, the proposed algorithm solves the recognition problems which are caused by rotation, scale, blur and illumination transformation, and achieves the requirements in real-time performance and industrialisation. It is demonstrated by experiment that compared with traditional component recognition algorithms, the new algorithm makes significant improvements in algorithm' s efficiency and stability.
出处
《计算机应用与软件》
CSCD
2015年第1期186-189,共4页
Computer Applications and Software
基金
国家自然科学基金项目(51207072)
关键词
SURF特征
零件识别
近似最近邻匹配法
Speeded-up robust features Component recognition Approximate nearest neighbour matching