| In recent years,with the development of artificial intelligence technology and the popularity of mobile devices,mobile services and applications have brought great convenience to people's work and life.Since walking is one of the most ordinary daily activities,the research of walking detection and step counting have attracted much attention in the fields of pedestrian localization and navigation,health monitoring,behavior recognition and etc.However,the current research has the following problems:(1)the use of the device is constrained,for example,the device can only be placed vertically in the trousers' pocket,reducing the practicality;(2)relying on dedicated equipment deployed in a specific location(such as dedicated foot acceleration sensor)not only limits the popularity of the application,but also increases the cost;(3)they does not consider various practical problems in the real scenes,such as varying walking speeds.In order to solve the above problems,this thesis deeply studies the walking detection and step counting methods using unconstrained smartphones in complex scenes.Firstly,unlike the existing general acceleration-based methods,this thesis uses the highly sensitive gyroscope to collect angular velocities,and extracts the frequency domain features of the smartphone carrier behavior in the angular velocities measurement by Fast Fourier Transform(FFT),and then based on the frequency domain features of the walking motion,a threshold-based walking detection algorithm is proposed.This algorithm reduces the constraints on how the device is used.Secondly,this thesis uses a weighted mean filter to smooth the real-time step frequency estimated by the polynomial fitting method,and then a novel FFT and Gyroscope(FG)step counting method is proposed;additionally,the Integral Step Counting(ISC)method and the Adaptive Sliding Distance Step Counting(ADSC)method are proposed for varying walking speeds to improve the practicability of the step counting methods.Finally,this thesis designs and carries out a large number of experiments in the real scenes to verify the effectiveness of the proposed methods;simultaneously,in the experiments this thesis considers the factors that affect the detection results,such as the placements of the mobile phone,the ground conditions and the walking speeds;the experimental results show that the proposed walking detection algorithm can achieve an average accuracy of 93.76% and an average recall rate of 93.65%,and the overall performance is significantly better than the standard deviation threshold(Standard Deviation Threshold,STD_TH)and FFT algorithms;compared with the commonly used peak detection,autocorrelation coefficients and the step counting software,the three new step counting methods proposed in this thesis have advantages in accuracy and stability.Among them the ADSC method has good performance,the average accuracy is as high as 95.39%,and the average standard deviation of accuracies is 3.85.In conclusion,the proposed methods can achieve high-precision walking detection and high-accuracy step counting in complex scenarios,which will play a positive role in promoting the application and development of walking detection and step counting. |
