Research On Moving Target Detection Algorithm Based On WiFi Through-the-wall Radar

With the advancement and development of science and technology,the need for seeing through walls is becoming more urgent,and through-the-wall target detection has extensive application prospects in security,smart home and emergency rescue.Due to the wide popularity of WiFi devices and the development of the capability of WiFi protocol,utilizing fine-grained channel state information to achieve through-the-wall target detection has attracted extensive attention from researchers.The existing through-the-wall target detection technologies based on channel state information contain following disadvantages: Firstly,relevant researches on clutter mitigation are lacking in the case that commercial WiFi devices are utilized to complete through-the-wall target detection.Secondly,the detection feature in most researches is single and the feature of frequency domain is less explored.Thirdly,the classification effect of the commonly used classifier depends on a few input features,which makes the detection accuracy low and the robustness weak.Aiming at above problems,a moving target detection algorithm based on WiFi through-the-wall radar is proposed in this thesis.The main contents include:Firstly,the clutter mitigation algorithm is studied.In this thesis,the WiFi through-the-wall radar platform is built by commercial WiFi devices,the received signal is analyzed,and the phase error is analyzed and corrected.Then,the received signal matrix is decomposed by singular value decomposition and the clutter subspace is determined by the maximum inter-class variance method.On this basis,the signal is reconstructed.Finally,the noise clutter in the reconstructed signal is removed by wavelet threshold denoising,and the signal after clutter mitigation is finally obtained.Secondly,the multi-feature extraction algorithm is carried out.The subcarriers which are more sensitive to the movement of target are selected.Then the correlation matrix of amplitude and phase between data packets are obtained by pearson correlation coefficient,and the maximum eigenvalues are extracted as the correlation features.For the problem that the received data is redundant,t-distributed stochastic neighbor embedding is utilized for data dimensionality reduction,and the time-frequency features are obtained by wavelet transform and statistical feature extraction.Thirdly,classifier is constructed based on random forest.In this process,feature extraction is carried out by assigning weights to features to make the extraction probability of imbalance features equal,so the multi-feature joint detection classifier is obtained which has strong robustness.Finally,data is collected in two different environments and relevant experiments are implemented.The experimental results show that the detection accuracy of the system can reach 80% of concrete wall and 90% of hollow wall,respectively.