Over the past few years, the use of Multiple Input Multiple Output (MIMO) radar has gained increased attention as a way to mitigate the degredation of micro-Doppler classification performance incurred when the aspect angle approaches 90 degrees. In this work, the efficacy of co-located MIMO radar is compared with that of distributed MIMO. The performance anaylsis is accomplished for three different classification problems: 1) discrimination of a walking group of people from a running group of people; 2) identification of individual human activities, and 3) classification of different types of walking. In the co-located configuration each radar is placed side by side so as to form a line. In the distributed configuration, the radar positions are separated to observe the subjects from different angles. Starting from the cadence velocity diagram (CVD), the Pseudo-Zernike moments based features are extracted because of their robustness with respect to unwanted scalar and angular dependencies. Two different approaches to integrate the features obtained from multi-aspect data are compared: concatenation and principal component analysis (PCA). Results show that a distributed MIMO configuration and use of PCA to fuse multiperspective features yields higher classification performance as compared to a co-located configuration or feature vector concatenation.
展开▼
