Corner features extraction: underwater SLAM in structured environments

Purpose – The purpose of this paper is to present a method to extract corner features for map building purposes in man-made structured underwater environments using the sliding-window technique. Design/methodology/approach – The sliding-window technique is used to extract corner features, and Mechanically Scanned Imaging Sonar (MSIS) is used to scan the environment for map building purposes. The tests were performed with real data collected in a swimming pool. Findings – The change in application environment and the use of MSIS present some important differences, which must be taken into account when dealing with acoustic data. These include motion-induced distortions, continuous data flow, low scan frequency and high noise levels. Only part of the data stored in each scan sector is important for feature extraction; therefore, a segmentation process is necessary to extract more significant information. To deal with continuous flow of data, data must be separated into 360° scan sectors. Although the vehicle is assumed to be static, there is a drift in both its rotational and translational motions because of currents in the water; these drifts induce distortions in acoustic images. Therefore, the bearing information and the current vehicle pose corresponding to the selected scan-lines must be stored and used to compensate for motion-induced distortions in the acoustic images. As the data received is very noisy, an averaging filter should be applied to achieve an even distribution of data points, although this is partly achieved through the segmentation process. On the selected sliding window, all the point pairs must pass the distance and angle tests before a corner can be initialised. This minimises mapping of outlier data points but can make the algorithm computationally expensive if the selected window is too wide. The results show the viability of this procedure under very noisy data. The technique has been applied to 50 data sets/scans sectors with a success rate of 83 per cent. Research limitations/implications – MSIS gives very noisy data. There are limited sensorial modes for underwater applications. Practical implications – The extraction of corner features in structured man-made underwater environments opens the door for SLAM systems to a wide range of applications and environments. Originality/value – A method to extract corner features for map building purposes in man-made structured underwater environments is presented using the sliding-window technique.