Elevation Estimation in 3D Radar by an Ensemble Regression Model for Surveillance Applications

Abstract

3D surveillance radars determine three main parameters: Range, Azimuth, and Elevation angle of the aerial target. The elevation estimation parameter is a key feature of 3D radars. Sequential lobing, conical scanning, and mono-pulse are the traditional methods to estimate the elevation angle of an aerial target. These methods have limitations in elevation estimation accuracy due to antenna pattern error, channel mismatch error, platform orientation, platform stabilization, jamming and clutter, multipath reflection, target fluctuation, etc. This paper suggests the machine learning based Ensemble Regression Elevation Estimation Method (EREEM) for elevation estimation in 3D radars. It is based on popular regression techniques such as Linear Regression, Decision Trees, Random Forest, Support Vector Regression, Gaussian Process Regression, Kernel Regression and Neural Network Regression. The accuracy of the proposed method is validated over simulated stacked pencil beam data as well as recorded data from 3D surveillance radars. It has higher accuracy over sequential lobing, conical scanning, and mono-pulse methods of angle estimation. Observed height accuracy is more than 95% using EREEM.