Research On Feature Extraction And Matching Algorithms Of AUV-SLAM

With the increasing of research status of the ocean, demand for underwater robots which have the ability of autonomous navigation and positioning, is becoming bigger and more demanding. In the field of robot, research on autonomous underwater robot AUV has become a hotspot. In the complicated and changeable underwater environment, the sonar and camera sensor as the "eyes" of AUV, shoulder the important mission of environmental perception. Also, the image analysis and processing has become an important issue in the field of AUV, and it determines the effectiveness and accuracy of subsequent navigation, localization and mapping.This paper combines the sonar sensor and monocular camera to collect images, and then analyzes, extracts and matches features, thus providing the environmental information that SLAM algorithm needs.For sonar sensor, this paper utilizes Sea ScanHDS side scan sonar to collect images. According to the characteristics of sonar images, features are extracted by image smoothing, thresholding and clustering analysis. Then, the normal distribution transform NDT algorithm is used to match features for relative-position constraints.For camera sensor, this paper computes the coverage of a pair of pictures based on link hypothesis before characteristic analysis, in order to determine the candidates and improve the efficiency of the algorithm. For specific applications, we apply a standard model of feature matching to match the candidates of image. In the feature extraction part, SIFT feature detectors are used within a pairwise image registration framework, and the PCCS algorithm is adopted to eliminate the wrong matches. In addition, a geometric model selection framework is employed to automatically choose the appropriate matching model to get precise information constraints in order to adapt the variability of the structure. Compared with previous algorithms, the method is more effective, real-time and accurate.