Research On The Method Of Close-range Image Orientation Based On Linear Feature Assistance

With the development of social economy,informatization and industrialization have been promoted.Driven by major policy guidelines such as “Digital Earth” and “Smart City”,how to locate 3D targets and collect real-time and accurate geographic data information has become the focus of photogrammetry research.Simultaneously,with the improvement of computer hardware and software,as well as the continuous development of pattern recognition,computer vision and other disciplines,digital photogrammetry is far beyond the scope of traditional photogrammetry,which brings new opportunities and challenges to photogrammetry.In recent years,the wide use of various small measurement platforms has gradually made close-range photogrammetry a hot research direction in the field of photogrammetry.After systematic theoretical research and a large number of engineering practices,close-range photogrammetry technology has become increasingly mature,and has been widely used in three-dimensional reconstruction,construction engineering,automobile manufacturing,medicine,ancient architecture and other fields.However,due to the limitations of these platform conditions,close-range images often have large viewing angles,large distortions,and unsatisfied overlap.The weak textures,repeated textures,and occlusions in the environment sometimes occur.In this paper,the orientation of close-range images is studied.According to the large number of linear features in close-range images,the line features are applied to the orientation assistance to increase the observation values in some scenes and improve the orientation accuracy and stability of close-range images through the form of line constraints.The main research work of this paper is as follows:1.The two popular line segment detection algorithms(LSD algorithm and EDLines algorithm)are studied and compared.After analyzing the principle and the existing defects,the gradient angle based LSD algorithm is used as the line extraction algorithm.Through the quadratic fitting and approaching merger of the detected straight line,the extraction result of the straight line segment is optimized,and the acquisition mode and accuracy of the straight line segment observation at the current stage are improved.2.The LBD algorithm is used as the basic algorithm of line matching.The line descriptors are studied.For example,the attributes such as geometry and gray scale included in the line features are used to filter the candidate sets of the matching lines using the generated descriptors and the primary geometric attributes.Aiming at the problems existing in the actual matching results,this paper further uses cross ratio,projection invariant and other advanced geometric attributes to calculate the matching,so as to obtain the best matching results and further improve the accuracy of the matching algorithm.3.This paper mainly focuses on the research of the algorithm of close-range image orientation based on linear assistance.In the case of the traditional point feature as the observed value,the linear feature observation is introduced,and the linear re-projection equation is further established according to the projection relation of the straight line.Then,through the method of vanishing point classification,the horizontal line and the vertical line are automatically classified,and the two special line constraints are incorporated into the absolute orientation and the difference process,and the adjustment model of the point line multi-feature joint is established to improve the orientation accuracy and reliability.Experimental results show that this paper can automatically match and classify image lines through straight line extraction,matching and vanishing point classification algorithms,which reduces the process of manual measurement.Line feature assisting has the effect of constraining and improving precision on image orientation results.Research significance and practical application significance.