Research On Person Re-identification Based On Pyroelectric Infrared Sensors

In many indoor scenes,such as homes and offices,people need to be identified,which is very important for smart home,intelligent security and other fields.Person reidentification is when a person is detected under a device and can be identified when he or she reappears.Traditional person re-identification technology uses camera-based system,which has high cost,huge computation and data throughput,as well as privacy issues.In a changeable external environment,it will be affected by Angle,clothing,background,light and other factors.Pyroelectric infrared(PIR)sensor as a kind of passive infrared detector,is only sensitive to infrared radiation of 8 to 14?m,does not involve the use of video image so that it will not involve privacy issues,and has obvious advantages in cost and power consumption,convenient installation and deployment,can well adapt to complex external environment,with the combination of wireless sensor network(WSN),deep learning technology,can realize the human body positioning,action recognition,person re-identification,etc.in indoor environment.Aiming at the node structure and node deployment of wireless pyroelectric infrared sensor network,the accuracy of person re-identification and the anti-interference stability of the system when dealing with noise interference,a series of studies are carried out.The main work includes the following aspects:(1)A new person re-identification system based on pyroelectric infrared sensor and distributed wireless sensor network is designed.Through the field of view modulation and deployment design of sensor nodes,the infrared signals of moving human body is obtained completely and accurately.Microcontroller unit(MCU)was used to complete A/D sampling of data,and Zig Bee wireless network protocol was used to transmit data to PC for signal analysis and processing.(2)A classification model based on CNN-LSTM deep learning neural network for processing infrared signal sequence data of persons is designed,and a motion feature representation method of different parts of persons based on the spectrum vector of FFT algorithm and a Concat feature fusion method are proposed.The collected original data is preprocessed on the PC,the signal spectrum is calculated to get the feature vector group,and the feature fusion of the vector group is carried out to get the complete features that can represent different people,and the classification and recognition of the feature data is realized through the deep learning neural network.(3)The possible noise sources of the system are analyzed,and a simulation of adding noise on the original data is carried out,and then a system implementation method based on signal denoising and deep learning is proposed,firstly,the wavelet threshold denoising is designed,and then the denoising method is used to denoise the noisy signal,combining with the designed deep learning neural network to realize the person re-identification under denoising data.The experimental results show that the system has good accuracy in the task of person re-identification in indoor scene,and has good ability of anti-noise interference.It provides a low-cost scheme for indoor person re-identification,and has good application value.