Research On Feature Extraction Algorithm Of ORB Image Based On FPGA

Image feature point extraction is an important part of the visual positioning algorithm.By extracting the features and realizing the mapping algorithm positioning,FPGA field programmable gate array technology improves the data throughput of the machine vision detection processing system due to its powerful parallel processing capability.Capabilities and data processing capabilities.In this thesis,the ORB image feature extraction algorithm is taken as the research object,aiming at extracting features with high frame rate without loss of positioning accuracy,focusing on illumination change adaptation,feature descriptor calculation,feature direction calculation,high efficiency and low overhead FPGA implementation.After key technologies are analyzed and studied in depth,the main work of this paper is as follows:1.Research on adaptive threshold FAST corner feature extraction method.The traditional FAST feature extraction adopts a single threshold and cannot adapt to a variety of images.Especially in the case of large illumination changes,the feature points extracted by the same scene will be greatly different.In this paper,the method of statistical analysis of image difference is adopted,and the image filtering is performed by small filter.The difference between the center point and the surrounding points in the filter kernel is statistically calculated and the statistical result is converted into a threshold adjusted within a certain range to improve the feature.Extract the effect.2.Research on feature direction calculation structure optimization.The traditional ORB feature direction angle calculation uses the gray scale centroid method.However,the calculation of the centroid vector angle algorithm is complex and takes up a lot of resources and clocks.Therefore,this paper divides the two-dimensional image plane into multiple sub-intervals and calculates each sub-interval.The interval function,and exponentially powering the variable coefficients of the interval function to remove the floating-point operation,and then directly substituting the obtained centroid coordinates into the sub-interval pairs to obtain the feature direction angle.3.Research on Visual-inertial SLAM algorithm.This study introduces IMU(Efficient and accurate study of the calculation method of the BRIEF descriptor descriptor.The traditionalRIEF feature algorithm consumes a large amount of resources and achieves a multiplication of the rotation direction.And the matching point is poorly robust.This paper proposes the method of fixed-point training and correlation test to improve the breadth of point pairing results.Descriptors are generated using a rotation mode to reduce the storage and logic consumption of the FPGA.4.Research on FPGA design methods for lightweight ORB feature extraction.Modular FPGA pipeline design for optimized ORB image feature extraction algorithm to achieve low-overhead and high-efficiency algorithm hardware acceleration target.