Implementation And Algorithm Of RANSAC Based On Image

With the development of science and technology,many computer vision technologies are playing an important role in social development.Digital image processing technology has been widely used in many fields such as industrial production,automatic navigation,environmental protection,climate monitoring,life science,terrain exploration,national defense and military.Currently,image matching method based on image feature extraction and RANSAC is the most widely used image matching techniques.This method uses the few interest points as image features,and then uses RANSAC to obtain the spatial mapping relationship of the matched images.Modern remote sensing,autonomous robots,intelligent monitoring and other fields all place high demands on the real-time performance of image matching technology.Software platforms are inefficient,and they are difficult to meet real-time computing requirements.They are bulky,costly,and the use of space is further reduced.With the development of semiconductor technology,many image algorithms have been designed to hardware structures that speed up the algorithm and ensure the real-time performance of the system.In this thesis,the projection transformation model,polar geometry model,camera movement model and image transformation model are introduced.The polar geometry model based on RANSAC image matching technology and several basic types of camera movement are analyzed.Then,this thesis introduces the most widely used image matching method based on RANSAC,and deduces the basic principle,processing flow and algorithm pre-processing of RANSAC algorithm.The mathematical model of the algorithm is deduced and improved based on the engineering requirements and hardware implement features.This thesis focuses on the analysis and derivation of the mathematical model calculation of the RANSAC algorithm based on the epipolar geometry,and advances the hardware design,mainly in the RANSAC algorithm sampling method,the homography matrix calculation process and the calculation of the number of iterations.Based on the improved algorithm,the hardware processing architecture of RANSAC algorithm is designed.The RANSAC hardware architecture and the sampling unit,the homography matrix calculation unit and the consistency comparison unit are analysed in four subsections.Based on the improved RANSAC method,the integral hardware implementation structure is designed.The RANSAC hardware structure is described using Verilog HDL language,and is simulated and tested on Mentor's Modelsim simulation platform.Based on the dual-view image simulation,the functional effects of the hardware structure are tested.After the RANSAC hardware structure is processed,the incorrect feature matching is basically eliminated.The remaining matching feature points show a unique mapping relationship between the two images.Finally,this thesis analyzes the timing of the RANSAC hardware structure and logic resource consumption.Compared with related research results,the RANSAC hardware structure designed in this thesis has improved both in algorithm performance and hardware resource usage efficiency.The RANSAC hardware structure designed and implemented in this thesis supports an image resolution of 2048×2048,a maximum number of feature points of 1024,and a processing speed of 55 fps.The required hardware resources are less than a quarter of other research results,and no external memory is required.