Research On Sensing Quality Enhancement Methods In Mobile Crowd Sensing Networks

Recently,with the rapid popularization of mobile terminals among urban crowd,large amount of human handled or vehicle mounted smart devices,are interconnected through mobile networks.Through intentional or unintentional cooperation,these smart devices can accomplish large-scale sensing tasks.This becomes a new sensing paradigm,known as "Crowd Sensing",for the Internet of Things(IoT).Compared with the traditional initiatively deployed sensor net-works,the crowd sensing networks can fully leverage the existing network and equipment resources,and flexibly accomplish diverse sensing tasks.However,because of the characteristic of "people-centric" sensing,crowd sensing nodes follow human mobility patterns,therefore their temporal-spatio distribution ex-hibits significant non-uniformity and dynamics.This leads to the existence of some "weak zone" even "blank zone" in the network monitoring region.How to effectively reduce the sensing weak zones and enhance the quality of crowd sensing has become an urgent issue to be solved.Thus,this thesis focus on enhancing the sensing quality of crowd sensing and proposes a serious of novel methods from three aspects:1)estimating the needed amount of users for data collection;2)fusing correlated sensory data;3)learning the structural infor-mation of sensory data for sensing quality enhancement.Specifically,the main contributions of this thesis are summarized as follows:(1)The supplementary node scale estimation method for crowd sensing networks.Enlarging the recruitment scale of crowd sensing node,is a way to enhance the temporal-spatio coverage capability of crowd sensing networks,therefore reduce the sensing weak zones.However,existing research shows that urban crowd has specific mobility patterns.So that the crowd scale and network temporal-spatio coverage capability exhibit a complex nonlinear rela-tionship.To address this issue,we first explore the evolution rule of temporal-spatio coverage quality of human crowd under different crowd scales,and fur-ther propose an urban context fused trace generation method.This method enhances the accuracy of estimating the needed supplementary node scale to cover the sensing weak zones.(2)Data correlation based crowd sensing enhancement method.Various of correlations widely exist in crowd sensing data.For example,in temporal-spatio dimensions,the temporal and spatial adjacent sensory data are corre-lated.Besides,different categories of sensory data are correlated.For mod-eling the data correlations on different dimensions jointly,we leverage tensor to model the crowd sensing data based on the sensing time,location and data category,and further propose a collaborative tensor decomposition method for sensory data fusion,in order to acquire the complementary data knowledge for enhancing the sensing quality in crowd sensing weak zones.(3)Data structural feature based crowd sensing enhancement method.Large amount of sensory data exhibit continuous variation in temporal-spatio dimensions,therefore have specific inner structural features.For the crowd sensing schema,because of the non-uniform distribution of crowd sensing n-odes,the crowd sensing networks face severe data missing in sensing weak zones,which brings difficulties in structural feature learning and the subse-quent missing data reconstruction processes.To address this issue,we first propose a data recovery error measurement method for the non-uniformly dis-tributed crowd sensing nodes,i.e.,impact map,then propose a nodes’ location fused generative adversarial network for data structure learning and mapping,through which the sensing quality in crowd sensing weak zones can be en-hanced.To evaluate the effectiveness of the proposed models and methods,we design and implement a prototype system for enhancing the sensing quality of crowd sensing networks.The extensive experimental results show that the proposed methods can enhance the sensing quality of crowd sensing networks.