| Simultaneous Localization and Mapping(SLAM) is the key technology of intelligent mobile robots. In recent years, SLAM technology, applied to the underwater environment, become a hot topic. however, there are many technical difficulties for underwater environment, owing to the presence of terrain, dimly light, much random interference and other factors. In view of these problems, the key technologies of visual SLAM system and image processing in the underwater environment was used as the research topic, focusing on the implementation of SLAM technologies, underwater image pre-processing problems and the algorithm of feature extraction and matching in the underwater environment. The main contents of this paper include the following aspects:Firstly, in this paper, vision-based SLAM algorithm was introduced and analyzed, and a mathematical model of EKF-SLAM system was built and simulated by MATLAB software. Get the difference between EKF-SLAM algorithm localization errors and odometer localization errors by simulation, then contrast the size of errors by drawing a map in the two localization methods. In this way, the accuracy and reliability of this EKF-SLAM method was proved, and a reliable EKF-SLAM system model was established.Secondly, the impact of the underwater environment on camera imaging was described, light refraction, scattering and absorption phenomena in the imaging process was analyzed and the corresponding solutions was put forward in this paper. Then, a dark pre-color image contrast enhancement method was introduced, which was applied in the foggy days, and then a dark pre-color contrast algorithm was proposed, which could be used in the underwater environment, and whose feasibility was proved by many experimental results, and provides an effective image enhancement method to the later feather extraction process.Once more, in this paper, after a general method for image feature extraction was described, the advantages and disadvantages and applicability in the case of the underwater environment of various methods was analyzed, the Scale-Invariant Feature Transform(SIFT) algorithm was verified to be a more appropriate feature extraction algorithm. Then, in view of the SIFT algorithm real-time performance and the shortcoming of redundancy feature points, a processing method based on the characteristics of the area was proposed, and whose good effect on the improving the real-time performance was testified by experiments, resulting in an improvement method of features extraction with high efficiency was concluded in the paper.Finally, the method of data association of SLAM system was analyzed, and whose importance status was demonstrated. In terms of the wrong matching and low matching efficiency problems in the data association process, in this paper, a solution was proposed, that is, to associate the relative location factors of binocular cameras with signs of location factors, as auxiliary conditions for data association. Experiments show that, after added the position information of landmarks, correlation feature complexity greatly reduced, especially with the increasing number of feature points, the time of matching feature points has been greatly shortened, the accuracy of which has also been improved and the real-time performance has been increased four times at the same time, bringing about a high efficiency method for data association. |
PDF Full Text Request