Target Detection By Marine Radar PDF Download

The growing energy needs have eventually increased the development of wind turbines. The constructions of wind turbines have several potential impacts of which the most significant factor is the increasing bird mortality rates due to collision and habitat loss. Since then, radars have been deployed to study the behavior of birds towards wind turbines. Radars employ target tracking for identifying the targets (birds) accurately and efficiently. Several methods of tracking were developed to improve the tracking efficiency of the radars over the years. Most widely used tracking techniques are Kalman filter and particle filter. These filters use data with random errors and estimate accurate values for the current state of the system. Kalman filter is a linear estimator which does not depend on a set of past observations and hence efficient in real time applications. The particle filter also known as sequential Monte Carlo method is a nonlinear estimator which uses a set of particles with various weights for estimation. However, particle filters have high computation time. Kalman and particle filters were developed over the years creating various models for various types of systems. A block version of Compressive Matching Pursuit (CoSaMP) algorithm used in signal reconstruction called BCoSaMP was employed in tracking. It was seen to give a similar or better performance than particle filter with less computation time. The BCoSaMP algorithm with Kalman filter estimation was developed which reduces the mean square error as compared to other models in certain cases. This thesis focuses on developing tracker models in radR. Kalman filter tracking model based on linear data and Gaussian noise that operates over a variety of target motions and velocities is developed. Particle filter is designed for nonlinear target motion with non-Gaussian noise. BCoSaMP model that assumes data as sparse is applied for target tracking and a modified BCoSaMP which replaces least square estimation with Kalman filter estimation are also implemented. These models were tested with different data sets and a comparative analysis is performed. The algorithms are tested on simulated data and marine radar data in radR to compare the effects of the developed tracker models with the conventional methods in radR. The hybrid algorithm is shown to have better performance over the other models in the case of simulated track for some targets. Particle filter has the highest detection rate with marine radar data.