Image Matching Algorithm Based On SIFT In The Vehicle Navigation

The aided inertial navigation based on Image matching is one direction of the combined navigation. SIFT algorithm is one of the most excellent algorithms in the realm of computer vision. This paper mainly studies how to apply the SIFT algorithm to the vehicle navigation effectively to locate targets accurately and stably.As the feature detection in SIFT is not effective enough to meet the demand of vehicle navigation, this paper proposes a fast detecting method based on improved Gaussian filter. Firstly, the Gaussian filter is decomposed to a series of weighted box filters; then fast weighted box filters are implemented based on integrate image, and the results are combined to realize Gaussian filter; at the end, the build of the scales-space can also be enhanced, which makes the feature detection can be parallel implemented. As the experiments show, the proposed method not only keeps the precision of the traditional SIFT, but also improves the efficiency about 30%.In order to overcome the unstable problem in the SIFT feature clustering to make the vehicle navigation more stable, we proposes an algorithm based on improved RANSAC. The algorithm first substitutes exhausted search for random sample in classical RANSAC; then combines the stable matching pairs and discards the unstable ones to cluster the correct matching pairs. As the experiments show, compared with the classical RANSAC, the performance of the improved RANSAC is more stable.