Research On Image Migration Motion Estimation Of Medical Infusion Based On Feature Matching

With the highly development of social economy and science and technology,industrial production has also gradually to the high speed and intelligent direction.But smart light inspection machine imaging system took the sequence of images taken in the bottle,can produce tiny movement deviation in horizontal direction and vertical direction.This movement deviation will lead to larger amount of pixel deviation on the image,the characteristics of the difference of the image processing will produce a lot of mistakes in the subsequent visual processing system,to a subsequent interframe visual processing difference for identification and judgment,increased the difficulty of foreign body,increase the misjudgment rate of the system.Based on the comprehensive comparison of various algorithms of motion estimation,and the comprehensive consideration of the image characteristics and real-time performance requirements of the application scene,the matching method of this thesis based on feature points was used to solve the motion vector.Considering the large amount of computation,the algorithm based on point cannot guarantee the real-time performance.In order to solve the problem,this thesis used the optimization of the improved version of the "FAST feature point detection + BRIEF binary descriptor" special combination in image matching,and use the matching point of the coordinate information to accurately solve the image motion vector offset.The shortcomings of FAST and BRIEF algorithm are improved and optimized.In this thesis,the threshold value of FAST algorithm is optimized,which not only eliminates a large number of false feature points,but also retains the majority of real feature points.At the same time,for the problem of noise sensitivity of FAST,the filter is used to reduce the noise influence,and a more stable feature point is obtained at a small time cost.The method of generating a BRIEF descriptor with more robust and stable to noise is generated by using the sub window instead of a single pixel.Finally,the matching accuracy is improved by using the optimal distance threshold and the block matching strategy.After obtaining the deviation data with strong stability and small error range,the histogram of the statistical deviation data is used to calculate the deviation of compensation.weighted the peak and auxiliary value of the statistical deviation histogram to solve the motion displacement offset vector.After contrast experiment simulation,the algorithm performance of this thesis is superior to the combination of original BRIEF algorithm and FAST algorithm,and it is superior to some existing classic algorithms show that this algorithm is feasible.