Research On Analysis And Recognition Technology Of Individual Behavior Features In Crowd-sensing

With the number of wearable and portable smart devices dramatic rising,and the types of sensors in these smart devices sharply increasing,a new perception of the Internet of Things named Crowd-Sensing was proposed.There are a lot of data about the user's behaviors in Crowd-Sensing.It becomes a new challenge that effectively using of the data and recognizing the behaviors features of individuals' and groups' in order to serve themselves.Nowadays,many studies of individual pattern recognition were proposed.Especially,the recognition based on acceleration is paid more attention to,because of the better applicability.However,the relative technologies are not good enough to classify many tiny or rapid behaviors.In order to further improve the performance of human motion behaviors recognition in Crowd-Sensing,this paper proposes a method using the sensor and no-sensor users' data generated by multi-types intelligence senses.This paper firstly optimizes the individual motion behavior recognition algorithm based on the acceleration data,and proposes a new feature extraction method based on the maximum spectrum FFT(MFFT)coefficient and the optimal feature subset selection(FFS)Multi-class classification algorithm,and effectively improve the recognition speed and recognition rate of human motion pattern recognition based on acceleration sensor.Secondly,based on the user's personal history motion data,a naive Bayesian classification algorithm is proposed.Based on the characteristics of individual users' behavioral habits,adaptive human motion pattern recognition classifier based on historical motion data is designed.And the k-nearest neighbor algorithm is used to design the human motion pattern recognition classifier based on the geographic position data.Experimental results show that the accuracy of the combined classifier is higher than that of the three sub-classifiers,and the variance of the combined classifier's accuracy is enormously lower than that of the each sub-classifier.