In this dissertation, novel approaches to achieve accurate underwater target tracking based on information fusion are presented. In practical applications of underwater target tracking, two types of platforms (e.g. sonar installed on moving platforms ...
In this dissertation, novel approaches to achieve accurate underwater target tracking based on information fusion are presented. In practical applications of underwater target tracking, two types of platforms (e.g. sonar installed on moving platforms and fixed multi-static sonar) must be considered. The first type, which could be referred to as a passive sensor installed on ships, bearings of targets are the only reliable form of acquirable information. Most conventional researches on bearing only target motion analysis (BOTMA) based on least squares (LS) have focused on performing asymptotically unbiased estimation with inaccurate measurements. Such researches achieve asymptotically unbiased estimation if a large set of measurements is obtained. However, in practice the demand for an immediate response in many maritime surveillance applications forces surveillance systems to make quick analyses of targets before sufficient measurements have been obtained. In the second type scenario, target tracking with a multi-static sonar network is investigated. Location of moving target often can be estimated using TDOA measurements from sensor network. Though, because the sensor network often obtains insufficient and inaccurate TDOA measurements due to the harsh underwater conditions, the performance of target tracking is easily degraded.
The purpose of this dissertation is to explore appropriate approaches to assure accurate target tracking under the aforementioned issues. In detail, for the first issue, an accurate target motion analysis from small measurement set by random sample consensus (RANSAC) algorithm is presented. In the BOTMA scenario, measurements with large errors are also regarded as outliers of the target motion model. Therefore, the purpose of applying RANSAC to BOTMA is to reduce outliers and enhance the performance of target motion estimation. For the second issue, track splitting algorithm for track management to solve insufficient TDOA measurements is proposed. Next, in order to accurately estimate target trajectory even in noisy underwater environments, a stack-based data association method is proposed. For performance assessment of the proposed methods, all experiments and analyses were conducted while promising results were attained with relevant underwater acoustics sensor network operational scenarios.