Image Matching Algorithm Based On Multi-scale Local Feature For Scene Matching Aided Navigation

Considering the requirement of scene matching aided navigation system, the situation of local feature scene matching algorithm is summarized, and the key technologies of scene matching aided navigation are analyzed. To overcome the weakness of feature point matching, a scene matching navigation algorithm based on multi-scale local feature is researched. In order to improve the suitability and precision, the steps of SIFT algorithm and the parameters influence rules to the match results are studied, the scene matching self-adapting navigation scheme and software flow are designed.In order to resolve the mismatch problem during the scene matching, an image space transform H model is constructed based on scene matching results. The efficient measure for solving the H model is designed by means of PROSAC algorithm, the scene matching results are evaluated and the mismatches are deleted with using H model transformation. A self-adjusting & self-assessment SIFT (S2SIFT) algorithm is proposed, the precision and reliability of the S2SIFT algorithm have been tested through lots of scene matching experiments.To enhance the real-time efficiency of scene matching aided navigation, a speeded up robust multi-scale local feature scene matching based on SURF is studied, the basic theory of SURF algorithm is expounded, the mainly speeded-up and robust technical characteristics of SURF algorithm are analyzed. Combined with the error characteristics analyzing of inertial device, revised methods for improving the matching speed of SURF algorithm have been proposed, which enhance the performance of speeded up robust scene matching algorithm.To check the effect of multi-scale local feature scene matching for inertial navigation, an integrated navigation scheme is designed based on Kalman filter, and a software emulation system is built based on VC++6.0. By means of analyzing the navigation result, the modification performance can be verified.The multi-scale local feature scene matching algorithm has high performance, which has great correctional effect on inertial navigation, the precision and reliability of the navigation system can be improved based on the algorithm in this paper.