AUTOMATIC RECOGNITION OF GROUND RADAR TARGETS

Abstract

The research work undertaken in this thesis is aimed at developing automatic target recognition (ATR) capability in Ground Surveillance Radar developed at the College of Electrical and Mechanical Engineering (E&ME), National University of Sciences and Technology (NUST), Pakistan. This work is based on the micro Doppler phenomenon. Three target classes i.e. vehicle, pedestrian and clutter have been considered. For this work, independent trials of the NUST Radar NR-V2 were conducted and target signatures were collected in a controlled environment. Joint time-frequency domain was selected for the purpose of analyzing the non-stationary nature of the micro Doppler signature of radar targets. A 2-D time-frequency minimum variance (TFMV) method has been used for obtaining high resolution time-frequency signature of the radar measured data. This method has been implemented using 2-D Capon spectral estimator which is a non-parametric spectral estimator. The resolution obtained with this method is greater than that obtained with linear time-frequency representations such as short-time Fourier transform (STFT) and is comparable to that obtained with the help of quadratic time-frequency representations (QTFRs), but is free of the cross-term artifacts which is a disadvantage of QTFRs. Alternatively, a classification approach has been tested on the radar data for the purpose of ATR. A novel three element feature vector has been used. Features are extracted from a 100 ms radar Doppler audio signal. The three features are: variance in target Doppler, average power of the radar audio signal and target range from the radar. The short feature length allows fast real-time implementation of the classifier. Classification is done using a k-Nearest Neighbors (k-NN) classifier. There are three input classes to the classifier; vehicle, pedestrian and clutter. Training has been done on real radar data. Classifier performance has been tested for the case of two targets (automobile and pedestrian), as well as for the case of three targets (automobile, pedestrian and clutter). For the first test case of two target classes, automobile is correctly classified 84.3% of the times and pedestrian is correctly classified 100% of the times. For the second test case of three target classes, the correct classification rate is 75.9% for automobiles, 93.5% for pedestrians and 96.3% for clutter.