Research On Outlier Detection Algorithm For Crowdsensing-based Internet Of Vehicles

By means of intelligent transportation technology,mobile Internet technology and Internet of Things technology,Internet of Vehicles(IoV)perceives the traffic information autonomously and achieves information exchange among vehicles,roads and the people.Furthermore,it provides drivers with a variety of transportation services.In IoV systems,traditional data acquisition methods have the disadvantages of high costs,small coverage and insufficient data volume.With the increasingly performance of smart mobile terminals(such as smart phones and tablets),a way to use this type of terminals for data collection,namely crowdsensing,comes into being.Using crowdsensing for traffic data acquisition can effectively derive vast amounts of traffic data at an extremely low cost.But at the same time,since most of the crowdsensing users are not trained or authorized,the "quality" of traffic data may deteriorate.In order to solve this problem,by utilizing the kernel density estimation(KDE)theory and the fog computing technology,this paper proposes an outlier detection scheme which is suitable for crowdsensing-based IoV.The main contributions in this paper are as follows:(1)A crowdsensing-based traffic data acquisition application software is developed.In order to imitate a crowdsensing environment and provide a reliable data derivation for the simulation of the subsequent detection scheme,this paper designs a traffic data acquisition application on the Android platform.In this application,the built-in sensors of the smart mobile terminals and the Baidu map module are used to collect and upload the traffic data in real time.In real scenarios,this application can gather traffic data at a very low cost.(2)An outlier decision algorithm for crowdsensing-based IoV is proposed.The traditional decision methods mostly use traffic flow theory or are based on a known data distribution to detect outliers.The detection performance is poor in crowdsensing-based IoV.Therefore,based on the analysis of the structure of crowdsensing data and the characteristics of outlies in crowdsensing-based IoV,this paper proposes an outlier decision algorithm based on KDE.And then we effectively modify the boundary bias problem occurs in KDE.Experiments show that this algorithm has better detection performance than traditional statistical decision algorithms.(3)An outlier detection scheme based on fog computing for crowdsensing-based IoV is presented.In the traditional cloud-computing-based detection scheme,the central servers have so heavy computation tasks that the detection time is too long.To solve this problem and make full use of the computing resource of the crowdsensing terminals,this paper introduces the fog computing technology to design a new detection scheme named as fog-computing-based outlier detection(FCOD).In FCOD,the outlier detection tasks are accomplished by the crowdsensing terminals.The detection scheme is divided into 3 phases:model initialization,outlier detection,and model update.Among them,model update is the focus of the scheme.This paper elaborates on the JS-divergence-based method for determining the significance of changes between the KDE models and the global KDE models'update method based on the forgetting mechanism.Experiments show that,compared with the traditional cloud-computing-based detection scheme,the proposed scheme can significantly save the detection time while the detection performance is almost unchanged.It can be inferred that the presented scheme can meet the needs of crowdsensing IoV in practice.