A Machine Learning Approach for Radar Based Height Estimation

This work proposes a new radar-based height estimation approach using machine learning. The approach takes advantage of multipath propagation of radar waves to produce robust features. A random forest regressor is trained using the obtained features for extended objects height estimation. First, data of extended targets with various heights located at different positions is recorded. Subsequently, range, angle, amplitude and phase information is processed using the Fourier based high resolution spectral estimation method RELAX. The processed information is then used to generate features based on a multipath target height equation which assumes the Line of Sight (LOS) and (Non Line of Sight) NLOS peaks are known. Finally acquired height estimation results from a trained random forest regressor are presented and discussed. The results show that the proposed method is capable of predicting target heights with high accuracy, without requiring the LOS and NLOS peaks to be known.