Fast Template Matching Based On Geometric Features

Machine vision is utilized to replace human eyes in completing the task of obser-vation and detection. It can not only improve the accuracy and efficiency of the detec-tion, but also be suitable to be used in the hazard and complicated environment. Templatematching is one of the researches in machine vision. Up till now, it is widely used in manypractical applications, such as target recognition, video tracking and precise positioning(e.g. die bonding, wire bonding etc) . As the upgrading of technology, requirements tomachine vision are higher and higher. It is of great significance to develop a stable, reli-able and accurate template matching algorithm, which satisfies the real-time requirementsof the hardware system.In this work, a fast template matching algorithm, based on the geometric feature ofthe image, is presented. Position and gradient direction of the edge points are used asthe matching information in calculating the similarity. Within this framework, in?uencesto the matching result from brightness variation of the image, non-uniform illuminationand image rotation are avoided. Moreover, in the case of partial occlusion, it is also wellapplied. In terms of principles in surface fitting, after applying a translation, an fast al-gorithm on producing the precise gradient and sub-pixel coordinates of the edge point isderived. The introduced algorithm speeds up the processing and improves the matchingaccuracy. In order to satisfy the real-time requirements, stopping criteria and image pyra-mid are employed to be the searching strategy. In the stopping criteria aspect, particularedge points are selected to be the test points. This operation improves the efficiency incomputing the matching score and avoids the fault judgment in partial occlusion whensingle stopping criteria is used. Furthermore, fast algorithm to get sub-pixel positioningand precise rotation is derived detailedly by least square algorithm with an approximation.Experimental results demonstrate that the algorithm introduced in this work is goodat disturbance and noise rejection. Reliability of the proposed template matching algo-rithm is not in?uenced by partial occlusion and illumination condition of the image. It isshown that the algorithm can also be applied to the target image with rotation with respectto the template. Moreover, within the condition of stability, reliability and high accuracy,the algorithm meets the real-time requirements.