Research On Aided Navigation Technique Of Optical Images In Ground Environment

Image aided navigation is inevitable tendency of integrated navigation system of vehicle.How to design an algorithm with high adaptability, precision and computing efficiency is the kernel study topic.Following the difficulties of IR/Optical images matching, this dissertation researches the key technologies of image aided navigation.The important contributions and creative achievements are summarized as follows:(1)By analyzing the character and difference of IR image and optical image,the edge types that exist in common area of images are described as mathematical models.Some common extraction methods are discussed and analyzed. A edge detection algorithm based on Tsallis entropy is presented. The algorithm uses local entropy information,and the divergence-direction is obtained by sliding a window over an image. The edge pixel is linked from candidate pixels using direction information. Some experimental results have been provided to show the edge detection capability of the proposed algorithm.(2)The edge-based image matching algorithms are studied. The algorithms include the similarity measurement based on 3-4DT, edge based similar distance(ESD),and the proposed similarity measurement based on distance transformation. Their performance on adaptability and precision are analysed and compared.The experiment results are also given. High computational complexity due to pixel-by-pixel search strategy is one major drawback. In this dissertation, a much faster algorithm to approximate the matching position between two images of a scene is described. Reduction on the computional complexity is achieved by using an the improved ACA in continuous space based search strategy,due to the non-exhaustive search nature of the algorithm.(3)The image matching technique based on fourier transform and Fourier-Mellin algorithm can be used to register images which are misaligned due to rotation, scaling and translation.It presents a method of image matching algorithm based on bispectrum, it uses the property that bispectrum is not sensitive to Gauss noise, which can exactly estimates the parameters between images suffered from translation, rotation and scaling distortion.(4)SIFT Operator has been proven to be the most robust local invariant feature descriptor,which has been widely thought of a success in image matching.The SIFT volumes of keypoints are reduced by kernel principal component analysis(KPCA).The coarse matching is established between two successive frames. A kind of highly robust estimation is presented,named least trimmed squares, on the data basis of the digital terrain map.The vehicle's position ,attitude and motion parameters is estimated by the proposed algorithm.