Method And Application Study Of Multi-source Spatial And Temporal Registration In Complex Low-altitude Environment

Multi source image registration is an important branch of image processing which has been developed in the field of spatiotemporal alignment with different resolution(multi granularity)and different physical characteristics(heterogeneous)data obtained by multisource imaging sensors in different imaging angles and different imaging times.The multisource data fusion of these heterogeneous data has the advantages of complementary information and overall display.However,the precondition of obtaining this advantage is multi source image registration,that is,multisource spatiotemporal consistency mapping.Multi source image registration technology can be applied to civil military and civilian applications,such as land surveying,security monitoring,ground monitoring,low altitude risk avoidance,emergency rescue,disaster prevention and disaster reduction,terrain matching,battlefield reconnaissance,marine maintenance,and so on,which can provide key technical support for improving the all-weather and all-weather environment perception ability in our country.Compared with homologous image registration,the research of multi-source image registration is not yet mature.Because of multi-source data heterogeneity,multi granularity and spatiotemporal inconsistency,the accuracy of multi-source space-time registration is low or even invalid.Therefore,it is of great significance to study the method of multi-source space-time registration in complex environment.The main work of the thesis is as follows:Multi-source heterogeneous and multi granularity data have the characteristics of large difference in feature and quality.There are many factors that cause interference image registration due to this feature,which leads to the high mismatching rate and even failure of the traditional multi source registration method using single feature measurement and constraint.To solve these problems,a multi-source spatiotemporal image registration algorithm based on multi feature joint constraints is proposed.The method is based on Harris corner detection algorithm to detect the corner point of multi-source spatiotemporal image.Using the multiple feature constraints of the location,distance and mutual information,the error matching is eliminated.The random sampling consistency algorithm is used for iterative fitting,thus reducing the error of the control point estimation,and finally accurately estimating the mapping matrix.The registration results of the Google optical image and the satellite borne SAR image show that the multi-source registration method effectively eliminates the mismatched corner pairs and achieves robust estimation of the multisource mapping matrix,thus achieving sub-pixel level registration accuracy.In view of the lack of feature points in some SAR images,such as grassland,desert,lake and ocean,the traditional feature based registration method is difficult to obtain the feature points effectively.Using the mutual information of mutual dependence between variables as an intelligent search algorithm for similarity measurement and geometric transformation parameters,the problem of lack of feature points is avoided.The mutual information is used as the similarity measure,and then the geometric transformation parameters are searched by intelligent method,and the difficulty of the higher mismatch rate of the more uniform scene registration is removed from the traditional feature based registration method.However,the existing intelligent search registration method based on mutual information has the defects of similarity measurement,poor robustness and local convergence.In order to improve the robustness of the similarity measure,the adjacent pixels are incorporated into the area of mutual information statistics to enhance the robustness of the similarity measure,and the whole search space is covered by the randomness of quantum,and the global optimal solution of the geometric transformation parameters is obtained by the quantum particle swarm intelligence search algorithm.Based on regional mutual information and quantum behaved particle swarm optimization,an intelligent search image registration method based on region mutual information is proposed.The results of the registration of Video SAR measured images in Sandia laboratory show that the method has the advantage of accurate similarity measurement and global rapid convergence,and achieves the registration effect of sub pixel level.