Automatic outlier detection from shallow water multibeam data using median filtering
Date
2003
Authors
Mann, Manjinder
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Accurate and fast processing of shallow water multibeam echosounder data is necessary to obtain high resolution bathymetric maps of a seafloor. The manual processing methods which are currently being used to identify the outliers in multibeam data, though accurate, are time consuming due to the large amount of data involved. In this thesis, an automatic outlier detection method is proposed to identify outliers from shallow water multibeam data fast and accurately.
Initially, two methods, robust estimation and median filtering, are described and implemented as possible candidates for an automatic method to detect outliers. Both methods are evaluated using synthetic data and based on the results obtained, median filtering is selected as the most promising candidate for automatic outlier detection. A new automatic outlier detection method is then proposed based on a two-stage median filtering algorithm. The method consists of three main parts, the preprocessing, the first stage, and the second stage of the two-stage median filtering algorithm. The preprocessing of the multibeam data is done to facilitate the implementation of a localization method used in the two-stage median filtering algorithm. The selection of parameters used in both stages of the median filtering is done based on the properties of the multibeam echosounder system and the multibeam data. Multibeam field data sets obtained from the Institute of Ocean Sciences (10S) are used to evaluate the performance of the proposed method. The processed multibeam data sets using the proposed method are compared with the results obtained by experienced operators at the 10S manually. The evaluation of the results indicate that over 95% of the outliers are detected and true objects on the seabed are preserved. The results are validated using visualization of the bathymetric images generated from the multibeam data sets.