Positioning System Based On Machine Vision Object Recognition

In today's society, automation degree of production line becomes more and more advanced. Using manipulator instead of traditional human is the trend. For manipulator, the precondition for gripping an object is to confirm the state and the position of the object. The system needs to identify and localize the objects by some method. In order to accurately identify and localize the objects, this paper is based on machine vision technology.This paper chooses a kind of motor as the study object. Firstly, obtain the motor images by the camera. Secondly, identify the motor in the images and determine the status of the motor and localize the motor by the computer. At last, calculate the angle and coordinate with the internal and external camera parameters. The main research contents and conclusions are shown as follows:(1)This paper chooses the motor images by the camera as the study object, and selects the appropriate method for this system after analyzes and comprises the image processing techniques and algorithms including the image grayscale, image enhancement, image binarization, edge detection and region filling.(2)Invariant moment is used in describing the shape of target and the feature extraction because it possess RTS invariability to images.After the image preprocessing and the feature extraction, the similarity is measured with characteristic parameters of the template image, and the angle and coordinate is calculated in the image coordinate system.(3)This paper uses the calibration method of Zhang, and sets up the camera model to calculate the internal and external camera parameters. The angle and coordinate of the motor are calculated in the world coordinate system with the internal and external camera parameters.(4)Using MATLAB designs this system, and using GUI displays the results of recognition. After analyzing experiment data, the recognition rate of this system is 90%, and the mean absolute error of the x direction is 5.9711mm, and the mean absolute error of the y direction is 4.4537mm, and the mean absolute error of the angle is 12.56°,and the y direction of the average absolute error is the average processing time is 11.4745 s. In conclusion, this paper shows this system could identify and localize the objects accurately.