Study Of Multi-camera Array System And Algorithm

Light information carries lots of clues about the scene,such as intensity,frequency,phase,angle etc.Ordinary imaging techniques or systems usually only focus on light intensity.However,in specific application fields such as remote sensing,life science,precision agriculture,material analysis,target recognition,etc.,it is necessary to perform a comprehensive analysis of the essential properties of the scene light.However,simultaneously acquiring multi-dimensional optical information such as high spatial resolution,high temporal resolution,and high spectral resolution of the scene has always been a core problem that plagues the industry.In recent years,thanks to the development of computational imaging technology,various computational imaging equipment and methods have been widely used in fields such as spectroscopy,polarization,and light field imaging.Among them,the multi-camera array technology has the advantages of low cost,high efficiency,and strong applicability compared with other existing imaging technologies,and has attracted more and more attention.This paper focuses on the application of multi-camera array systems to the collection and reconstruction of spectral and polarization information.The main work and innovations are as follows:(1)An image registration algorithm based on iterative optimization and reprojection is proposed,which is applied to the acquisition and reconstruction of spectral information of multi-camera arrays.Through pre-calibration and the introduction of enhanced correlation coefficients,the problem of conventional feature point registration algorithms that are difficult to find robust feature points due to the large difference in grayscale and texture between different spectral channels is solved,and the robustness and computational efficiency of the system are improved.Meanwhile,a target recognition algorithm based on a multi-spectral probability model is proposed,which realizes online target recognition and tracking at the megapixel level on the embedded platform.(2)An image registration algorithm based on depth map is proposed,which is applied to the acquisition and reconstruction of polarization information of multi-camera arrays.Conventional registration methods using homogeneous transformations are subject to inherent model defects,and the error increases when the scene depth differs greatly.For this reason,the convolutional neural network is introduced to calculate the scene depth map as a registration basis,which improves the accuracy of spectrum and polarization information collection and reconstruction of multi-camera arrays.(3)Based on the embedded system platform,a set of multi-camera array data acquisition and processing interface SDK is developed,which provides a convenient secondary development interface.At the same time,the SDK is used to build a multi-spectral camera calibration tool and a multi-spectral realtime acquisition and analysis software,which integrates camera calibration,spectrum correction and registration algorithms,improves the efficiency of the image registration algorithm pre-calibration process,and makes the multicamera array platform ease of use and more interactive.