Real‐time recognition of human motions using multidimensional features in ultrawideband biological radar

Abstract

Human motion recognition for biological radar has made astonishing progress. However, in some applications with high real-time requirements, it is difficult for existing approaches to achieve high accuracy. A multidimensional features long short-term memory (LSTM) neural network model is presented using multibranch network structure and high-dimensional radar feature fusion, which can recognise motions of human in real time, even in the presence of occlusions. The features selected for motion recognition including slow time range-map and slow time Doppler map. A single feature-based representation is not enough to capture the variations and attributes of individuals (range, velocity, etc.); thus, the fusion of multiple features is significant for recognising motions. Furthermore, because action reflects the behaviour of a human within a period, and the start and end are unavailable, intercepting fixed-length data in the time domain for recognition is not feasible. Thus, we introduce an approach based on an LSTM network that extracts features along the time dimension. Experiments show that the proposed approach is effective. A recognition accuracy of above 93.38% is achieved.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

Target Perception and Behavioral Recognition Algorithms Based on Saliency and Feature Extraction

“…(14) i.e. the intra-class variance can be obtained as (15) The inter-class variance is 𝜎 (𝑇). (16) Finally, using normalisation, we can obtain: (17) By combining the intra-class variance as a thresholding criterion for image segmentation, we can effectively separate the salient classes.…”

“…By performing a simple segmentation, regions of the human body pose such as feet, hands, head, etc. together determine the final matching result, where each part has its own weight [15][16][17].…”