Research On Method Of Hiden Human Targets Behavior Recognition

To correctly recognize covert human behavior is very significant to anti-terrorist and medical monitoring.This paper focuses on the modeling and preprocessing of human radar echoes,the construction of neural networks to recogonize human behavior in realtime,and the evaluation of real-time behavior recognition.The specific work is as follows:1.To solve the non-real-time identification application problem of the traditional human behavior global description model,the discretized human behavior description model is constructed firstly,and the local feature description of the human body behavior is realized,which provides a model basis for shielding the human body real-time behavior recognition.Then,set up the laboratory environment to collect actual human radar echo data.Secondly,based on the proposed human behavior model,the radar echo data is tagged,normalized,batched and out of order,which provides a data foundation for the realization of the real-time recognition model of the subsequent masking human behavior.2.To solve the real-time recognition and processing of high-dimensional timevarying non-equal length obscured human behavior data,a behavioral data reduction network based on fully connected neural layer is established,which effectively extracts low-dimensional short-term local features that obscure human behavior.Secondly,it establishes Based on the long-term memory unit,the real-time recognition network for masking human behavior has the ability to make short-term judgments of human behavior types.Finally,the BPTT training method based on random segmentation of behavior data is studied to realize the real-time recognition of masking human behavior.3.To solve the problem of reasonable evaluation of the real-time recognition model for masking human body behavior,the evaluation criteria of real-time recognition model based on IOU-F1 Score were studied,and the real-time recognition and recognition stability of the masked human behavior recognition model were accurately evaluated.4.To solve the problem that the real-time recognition model of masking human body recognition is not stable,a recognition robustness enhancement network based on long and short time memory unit is constructed.Under the premise of ensuring the real-time performance is basically unchanged,the stability of the model behavior recognition result is enhanced and realized.Real-time robust identification that masks human behavior.The above method was verified by the measured data.