Human Target Tracking in Multistatic Ultra-Wideband Radar

Abstract

Multistatic ultra-wideband radar is acknowledged to be useful for human tracking in indoor surveillance. In this thesis, a global signal processing chain for detection and localization of moving human targets using a multistatic ultra-wideband radar system is proposed. The design options for the system are presented, and the required signal processing steps are summarized. They include critical steps of tracking, target feature extraction,and target association. A decentralized tracking approach is designed for indoor surveillance. The 1D bistatic tracking rejects clutter and multipath significantly. The 2D tracker employs the global nearest neighbor based Kalman filter to track targets in Cartesian coordinates. The algorithm performance is compared with that of the conventional centralized approach on both simulated and experimental data, and shown to improve clutter and multipath rejection. For feature extraction, the advantages of using range-Doppler imaging are analyzed. The hypothesis-testing based and the Keystone transform based range migration compensation approaches are extended to eliminate the Doppler widening effect in range-Doppler processing of ultra-wideband radar. The proposed compensation approaches are applied to both simulated and experimental data. Special attention is then devoted to the critical issue of target association. Two slow-time features, video time density function and self-similarity matrix, are proposed to characterize target responses in different receiving channels. They indicate the slow-time evolution characteristic of the moving human targets, while staying invariant to radar observation angles. Auction algorithm based global nearest neighbor approach is proposed to tackle the measurement-to-measurement association issue. The decentralized tracking approach is experimentally verified on measured data for indoor detection and location of multiple human targets. Finally, as a by-product of human target feature analysis, range-Doppler surface is proposed to demonstrate the overall target information in range-Doppler video sequence. The conclusion emphasizes the main achievements and highlights the novelties. The recommendation for future work is presented finally to point out potential research topics related to this thesis.