Study Of Monocular Vision Target Recognition And Location Based On SIFT Algorithm

With the coming of industrial 4.0,automated production will reach a new level.One of the keys to achieving this goal is the development of intelligent robotics.Visual technology can simulate biological eyes.By analyzing the visual images,we can accurately perceive and understand the three-dimensional world.This is one of the key technologies to make the robot more intelligent.Therefore,the study of robot vision technology has an important theoretical value and practical application significance.This paper will use monochrome vision to obtain images information,and identify the target object by the improved SIFT algorithm.Than calculate object's coordinates in space by the monocular localization algorithm.At last,controlling the manipulator to grasp the target.This paper's main works as follows:Study camera's linear and nonlinear imaging modes.Analyzing Zhang Zhengyou camera calibration algorithm,and taking camera's distortion into account to reason the camera calibration principle.Doing the camera calibration experiment by OpenCV programming.The experimental results show that Zhang Zhengyou calibration algorithm can complete the camera calibration well,and get the camera's internal parameters and distortion coefficient.Considering the scale invariant feature transform(SIFT)algorithm's shortcomings and the application scenario in this paper,propose an improved SIFT with adaptive threshold and simplified descriptor.At the matching stage,proposing an matching method base on two-threshold similarity measurement and removing wrong matching points by Mahalanobis distance.Using improved SIFT algorithm to recognize object,the experiment carries out by OpenCV programming.The experiment result shows that the improved SIFT algorithm has better robustness to the illumination than the traditional SIFT algorithm,and the running time of the algorithm is reduced,and the correctness of the matching is improved.According to the application scenario,the localization algorithm used in this paper is based on a single image.Combining the four markers on the operating platform of the manipulator and camera's known internal parameters and distortion parameters,the relationship can be calculated between the image and the operating platform.Than according to the location of the target in the image,the coordinates of the target on the operating platform can be obtained,and finally the object positioning based on monocular vision is completed.At last,taking the KUKA manipulator as the experimental platform,the improved SIFT algorithm and the monocular localization algorithm are used to complete the crawling of the target object.