Abstract: It is necessary to identify and locate the target body quickly and accurately before grabbing the target object in the intelligent cigarette packing palletizing system. The traditional SURF method has a long time to detect the feature points. Therefore, a new method of binocular target matching based on improved FAST-SURF algorithm is proposed. In this paper, the number of detection pixels of the FAST algorithm is reduced from 16 to 12, and the matching feature points are searched by FLANN algorithm, and the matching time is shortened. Finally, the mismatch pair is eliminated by RANSAC algorithm, and the regular distribution of the internal and external parameters and the matching feature points obtained by the image matching results will be approximated according to the results of the image matching. The three dimensional coordinates of all the feature points are averaged and the approximate plane centroid coordinates are obtained. In the experiment, the threshold of the algorithm is discussed and the appropriate threshold of the algorithm is obtained. The experimental results prove that this method improves the speed and accuracy of cigarette package identification and has better robustness.

Key words: palletizing system, feature points, detection FAST-SURF algorithm, target matching, RANSAC algorithm

摘要： 在智能卷烟包件码垛系统中抓取目标物体前需要快速而准确地对其进行识别与定位。而传统的SURF方法在进行特征点检测时时间较长。因此，提出一种基于改进的FAST-SURF算法的双目目标匹配定位方法。将FAST算法检测特征点的检测像素个数由16个降为12个，利用FLANN算法搜索待匹配特征点，缩短搜索匹配时间，然后用改进RANSAC算法剔除误匹配对，根据图像匹配的结果与标定得到的内外参以及匹配特征点近似均匀分布的规律，将所有特征点的三维坐标进行平均运算，得到近似形心坐标。实验中，确定了合适的阈值，证明了该方法在卷烟包件识别定位的速度和正确率上都有一定提高且具有更好的鲁棒性。

关键词: 码垛系统, 特征点检测, FAST-SURF算法, 目标匹配, RANSAC算法