Research On Autonomous Navigation Method Of Underwater Vehicle Based On Multi-sensor Fussion

As the exploration and development of ocean progresses, the demand of underwater vehicle with independent navigation capabilities is growing. In the complex underwater environments, single sensor device, such as sonar, camera, etc., is unable to meet the requirements of the autonomous navigation, and the method of multi-sensor fusion becomes a best choice. After studying the simultaneous localization and mapping algorithm, this essay guarantees a broader, more long-term, more complex work area, and then will significantly advance our underwater autonomous navigation technology.This eaasy combines the current autonomous navigation technique, describes technical difficulties that autonomous navigation often encountered. After research on the two popular filtering methods, extended Kalman filter and particle filter, it provides several improvements based on the two methods to improve the precision and reliability of the algorithm.Fisrt, in the extended Kalman filter, because of the complexity of underwater environment, this essay uses sensor fusion of the sonar, camera, structured light to guarantee the validity of the observation data and information richness. Second the extended Kalman algorithm makes use of the update of the covariance matrix. As time passes, the number of the environmental characteristics increases, then the algorithm computation becomes difficult. In order to extend the duration of the algorithm, this essay introduces a temporary feature map. It reduces the growth rate of the number of map features, and reduces unnecessary computation effectively. Finally, the algorithm will divergence if there is an error of data association. To deal with it, this essay introduces a global scan matching algorithm based on data collected from scanner, which improves the efficiency and accuracy of data association.In particle filter, this essay first analyzes the problems in the filtering process, particulately the sample impoverishment. Secondly, this essay examines the roots of the problem, namely that the resampling process only retains particles with larger weight, and wipes off particles with smaller weight. Then a new particle filter based on particle splitting is proposed. The new method introduces particle preprocessing before resampling, namely particle splitting. The preprocess splits particles with larger weight into several particles with smaller weight, which obeys the same distribution. And then by resampling from the new particles set, the sample impoverishment problem is avoided.After the improvement of these two kinds of filtering methods, this essay simulate both methods. The experiment's results show that the improved algorithm can be good enough to achieve the desired objectives. It improves filtering accuracy and reliability, enhances practicality of filtering.