Research On Movement Detection With Unconstrained Smartphones

In recent years,with the development of artificial intelligence and the popularity of wearable devices embedded with various inertial sensors,the research and application of behavior recognition are growing rapidly.As an important branch of behavior recognition,pedestrian movement detection plays an important role in many fields,e.g.equipment localization,energy efficiency,medical care and etc.However,present research is confronted with two key issues.Firstly,dedicated sensor devices incur additional costs.Secondly,strict constraints on the placement of sensor devices poses limitations on applications.Meanwhile,the growing popularity of smartphones embedded with accelerometers,gyroscopes and magnetometers makes it promising to detect the pedestrian movement via smartphones.This thesis proposes a pedestrian movement detection method by using gyroscope measurements obtained by the smartphone carried by the pedestrian at arbitrary position.Specifically,by using the fast Fourier transform(FFT),the frequency-domain features in angular velocity are efficiently extracted in the first step.Then,after deeply analyzing the features of a moving pedestrian with an unconstrained smartphone,a threshold based model is established.Finally,extensive experiments are carried out in real scenarios,and a thorough comparison is made between the proposed method and another two popular methods,i.e.the standard deviation threshold(Standard Deviation Threshold,STD_TH)method and the short-time Fourier transform(Short Term Fourier Transform,STFT)method.The experimental results indicate that the successful rate of the proposed method is between 83%and 92.66%,whereas those of STFT and STD_TH are as high as 79.40%and 67.80%,respectively.Hence,the effectiveness and feasibility of the proposed method is then confirmed.In conclusion,the proposed movement detection method has demonstrated obvious advantages over other existing methods in terms of placement constraints and detection accuracy,and will evidently contribute to the applications and research based on behavior recognition.