Research On Continuous Casting Slab Model Size Measurement Based On Feature Point Detection And Matching

There will be burrs on the surface of continuous casting slab after hot rolling in the ferrous metallurgy industry.The traditional deburring methods are mechanical and flame,however,there are some problems such as insignificant deburring effect.With the development of technology in recent years,the method of cutting burr by plasma robot has been applied.In order to better assist the robot in deburring,it is necessary to complete the positioning and measurement of the continuous casting slab with binocular vision.Based on the Open Fund Project of Anhui Province Key Laboratory of Special Heavy Load Robot(TZJQR005-2021)and the Natural Science Research of Anhui University(KJ2020A0259),the measurement method of continuous casting slab size based on binocular vision is studied and validated by the continuous casting slab model.The specific research contents are as follows:(1)The camera calibration method and binocular vision measurement principle were studied.The two camera imaging models,binocular vision measurement principle and camera calibration were introduced at first.Then,the calibration of the binocular camera was completed based on the Zhang Zhengyou’s method.Finally,the calibration results were obtained.(2)The feature point detection methods were studied.The feature point detection algorithms,such as SIFT,Harris and FAST,were introduced in detail.However,the SIFT algorithm has disadvantages of the long time to detect and inaccurate detection;the Harris algorithm has the problem of missing detection;the FAST algorithm has the problems that the threshold value depends on artificial determination and the clustering corner phenomenon exists.Therefore,the improved FAST algorithm was proposed,that is,the detection template was simplified and the threshold value was determined based on the image contrast.Finally,the feasibility of the improved algorithm was verified by experiments.(3)The feature point description and stereo matching were studied.The construction method of SIFT descriptor based on gradient histogram and the r BRIEF descriptor based on binary were introduced respectively.The SIFT algorithm has a high matching rate due to the complex descriptors structure,which also leads to a long matching time yet;the r BRIEF descriptor has a short matching time due to the simple construction,but the matching accuracy is low.Thus,an improved r BRIEF descriptor was proposed,that is,the larger value of horizontal and vertical gradients was used to replace the traditional grayscale value for comparison,and the descriptor was formed.The Hamming distance was used to complete match and the mismatching was removed based on the epipopar constraint method.The experimental results show that the improved r BRIEF descriptor has a great improvement in the matching accuracy.(4)The measurement experiment of continuous casting slab model size based on binocular vision was carried out.Firstly,the improved FAST algorithm and the improved r BRIEF descriptor were used to detect and describe the feature points respectively.Then,the Hamming distance and epipolar line constraints were used to complete the matching,and the key corner points were obtained after screening.In order to improve the measurement accuracy,the sub-pixel coordinates of the key corners were obtained by the gray gradient method,and the internal and external parameters were used to obtain the three-dimensional coordinates information of corner points.Finally,the dimensional measurement work was completed.It was found that the measurement accuracy can meet the requirements according to the comparison between the real size of the continuous casting slab model and the measurement results,which verified the feasibility of the improved algorithm.