| In the robotics,underwater environments detection and objects feature extraction are the kernels of the AUV research and the important parts of the artificial intelligence study. In order to achieve the capability of autonomous tasks in an unknown environment, AUV should be capable of finding itself and recognizing objects from the environment. For the complicated and changeful condition, sonar system is the main detection equipment and composes the chief part of AUV's visual sensor. Different from the traditional ones, sonar system of AUV should have the functions of perceiving environment, obtaining information, building sonar images and extracting objects'features.Our researches are based on the National High-tech Research Development Plans (863 plans): AUV (Autonomous Underwater Vehicle) Navigation and location bases on simultaneous location and map building. This paper uses Seaking DST which is mechanically scanned forward-looking imaging sonar to scan and detection underwater environment and obtains the echo signals. After effectively selecting and processing the signals, sonar system can build a series of continuous sonar original image frames and underwater environment models, and then process these frames and extract objects'feature.In this paper, features extraction is carried out for tow purposes: one purpose bases on SLAM which only extracting interested point-features after the original sonar image frames preprocessing and these environmental points will be added to the continuous updating feature map, using SLAM algorithm to implement autonomous navigation of AUV. The other purpose is to advance the capacity of autonomous identification. After preprocessing, objects'geometrical characteristics become the most useful information. This paper uses least-square fitting theory and ellipse-fitting method to classify and extract multi-features. The result indicates that these approaches enhance the resolving power and discernment of AUV's visual system. |
PDF Full Text Request