Research On WiFi-based Indoor Passive People Counting Technology

In many scientific and technological research fields,WiFi-based indoor passive people counting technology has been highly valued by domestic and foreign researchers due to the natural advantages of indoor wireless networks and the important value of its own technology.The existing WiFi-based indoor passive people counting technology mainly uses the multi-classification model which is built in the training phase to realize online people counting.However,this type of technology has some disadvantages such as poor real-time performance,high training cost,and limited recognition range.In order to cope with the shortcomings of the above-mentioned people counting technology,the research of indoor passive people counting technology based on channel state information is carried out in this thesis.The specific research contents are as follows.First,the research on human activity detection algorithm is carried out.Firstly,on the premise of the difference in signal correlation between the unmanned silence and the human activity scenarios,the maximum eigenvalue after normalization of the correlation coefficient matrix is used as a feature to describe the signal difference,and the feature is extracted by sliding window.Then,the kernel density estimation algorithm is used to analyze the silent characteristics to obtain the best threshold.Finally,the adaptive threshold is updated through accumulating the silent characteristics of the online phase.The algorithm is tested and verified in multiple indoor scenes,and the results show that the algorithm can reach a higher detection level under the condition of fewer training samples.Meanwhile,the introduction of the algorithm proposed improves the recognition performance of the people counting system and reduces the computational complexity.Second,the research on passive people counting algorithm is carried out.Aiming at the problem of the single feature of the traditional people counting algorithm,the multiple features of time domain and frequency domain are combined to form a feature matrix to describe the number of active people in the environment.Besides,aiming at the problems of slow speed and premature convergence of traditional support vector machine,the parameter optimization based on the improved genetic algorithm is studied and a passive people counting algorithm is proposed.According to the experimental results in the real environment,the passive people counting algorithm designed in this thesis not only can realize fast and automatic parameter optimization,but also achieves a high accuracy of people counting recognition.Third,the research on passive population prediction algorithm is carried out.With the increase in the number of active people,the characteristic of the changes in indoor wireless signals shows a certain trend.Combined with the support vector regression algorithm,a reliable population prediction model is established to predict the number of detected people beyond the training range.The algorithm is tested and analyzed in a real indoor test environment,and the results show that the addition of the passive population prediction algorithm can solve the actual problem that the traditional passive people counting algorithm can not estimate the number of people which is outside the training range with a higher recognition accuracy.