Fast Model Matching Based On Feature Points And Direction Vector

With the rapid development of visual sensors and the maturity of a variety of computer vision algorithms, more and more machine vision inspection methods have been applied to the field of industrial inspection. Visual inspection method,which is non-contacting, high speed, high precision and highly automated, can make people far away from the dangerous workplace, and improve the production efficiency greatly, suitable for target location and detection of large quantities on the assembly line. Traditional object matching algorithms usually only take into account either the matching accuracy or the speed.This paper presents a hierarchical matching method. Firstly, based on ORB feature points, the extracted feature points in the image are matched with those in the template using the Hamming distance. Random Sample Consensus is used to filter most of the uncorrected matching and obtain the affine transformation parameters.Then, the orientation and relative location of each matched feature point pair were used to estimate the geometric center of each duplicate. K-means was used to cluster the center points obtained, thus that feature points belong to different duplicates of the object can be classified.Lastly, an improved direction vector extraction method is proposed, which is used to compare the edge information between the template and the image. In order to accelerate the calculation, the image pyramid algorithm is adopted, which locates the object from coarse to fine spatial scales. To obtain a subpixel-level matching result, a polynomial fitting method is used. The subpixel position and rotation angle are obtained by seeking a third order surface’s local maxima.Experiments show that the proposed method can separate the multiple duplicates of the target in the image, with high matching speed, good stability, and robustness to rotation and illumination.