Abstract:
Human targets display typical properties of multiple scattering relative to radar, and the strong scattering echoes from the torso often conceal the weaker scattering echoes from the limbs, head, and other body parts, thereby limiting the performance of activity recognition. To address this issue, a human activity recognition method using multi-channel radar based on scattering separation is proposed; firstly, the echo data of humans from multiple transmit-receive channels is stacked, followed by principal component analysis to separate the strong scattering echoes from the torso and the weaker scattering echoes from the limbs, head, and others into the first two components, which helps in avoiding the interference caused by concealment effects. Subsequently, short-time Fourier transforms are conducted individually to obtain time-frequency spectrograms corresponding to trunk movements and limb-head movements, expressing the characteristics of human activities jointly; after that, the histograms of oriented gradients for the two spectrograms are calculated, and the two features are then combined to form human activity characteristics, which are inputted into a support vector machine for recognition. Datasets encompassing six activities are gathered using a two-transmit-four-receive stepped-frequency radar. Test results demonstrate that this proposed method enhances the average recognition rate by 4.26% compared to approaches without scattering separation, and features of activities are fully expressed, which provides a new idea for human behavior recognition.