Research On Several Key Technologies Of Collaborative Computing For Crowd Sensing Systems

Crowd sensing can be divided into two scenarios:passive crowd sensing and active participatory sensing.In the scenario of passive crowd sensing,the carriers of the sensing devices are buses,unmanned vehicles and drones.In this scenario,the challenge is:how to ensure the accuracy of the data collec-tion while sense the environment in large and fine-grained scale.Especially,when the carriers are vehicles or drones,how to restore an air-pollution map that is with higher resolution and accuracy but generated from the limited ob-servation data,is also a challenge.In the active participatory sensing where the user participates,the carrier of the sensing device can be a smart phone or a wearable device such as an iPad,google glasses,a smart watch,etc.In this scenario,the number of data samples is usually limited,it is hard to extract a generic model from the collected sensing data in small sample scenario.How to learn a general model in small sample scenario is a critical issue that needs a solution.In addition,in this scenario,how to attract more participants to partic-ipate in the perception system is another challenge.A lot of previous work have studied how to attract more participants through incentives.However,when a task consumes too many power and computing resources of user devices,it may significantly affect the willingness of the users to participate in the task,even if there is incentive for the users.In order to solve the above issues,this thesis has done research on the two scenarios:the scenario of active participatory sensing,and the other scenario of passive crowd sensing.The contribution of this paper can be summarized as the following four aspects:Firstly,to provide real-time and accurate data covering large area,this pa-per proposes a novel scheme that jointly considers online multi-hop calibration and spatio-temporal coverage in route selection for mobile sensors.And a novel sensor carrier selection problem(SCSP)is formulated.An online Bayesian based collaborative calibration(OBCC)algorithm and a multi-hop calibration judgment algorithm(MCJA)are proposed.Based on the OBCC and the MCJA,a heuristic sensor route selection algorithm(SRSA)is then developed to solve the SCSP.We have conducted extensive simulations using real air pollution data and bus routes to evaluate the performance of our proposed solution.The simulation results show that,compared with traditional approach,our approach can reduce 60%of the original mobile sensors and can acquire much higher spatio-temporal coverage ratio and data accuracy.Secondly,in order to restore highly resolved and accurate air pollution maps,which are valuable resources for many issues related to air quality in-cluding exposure modeling and urban planning,this paper have studied how to use the spacial-temporal correlation of air pollutants to the rebuilt missing data in the case of small sample,so as to realize high resolved and large scale fine-grained air pollutant monitoring.To improve the accuracy of data rebuilt,this paper proposes a novel scheme that jointly considers sensor calibration and data rebuilt in route design for mobile sensors.We formulate a novel sensor route planning problem(SRPP)and propose a heuristic algorithm to solve the SRPP.The simulation results show that,compared with traditional approach,our approach can reduce 83%root mean square error(RMSE)on average.Thirdly,to solve the problem of how to learn the general model through small sample and complete the migration learning,we build an IoT-based group-aware platform system that continuously collects 16 months of photo data and related heterogeneous data such as camera lens parameter data,shooting posi-tion and angle information.We conducted an in-depth study of the correlation between collected data and air pollutants,and established an image based in-telligent station(IBIS)based on pixel-wise depth of images.We divide IBIS into two categories,one is the learning IBIS;the other is transferring IBIS.The learning IBIS is built nearby the air quality monitor station and the images collected by the learning IBIS have same shooting angle.The transfer IBIS has no restriction on the shooting angle and location,where the photos can be taken anyway.The learning IBIS builds cloud-based knowledge.According to the knowledge,the transferring intelligent can estimate the air quality con-centration without learning.Then,this paper has evaluated the performance of migration learning under the real data set.The experimental results show that the migration learning is well implemented in the small sample scenario.Finally,to solve the problem that the perceived task consumes too much power and computing resources of user devices,this paper proposes a general collaborative computing architecture where the users are grouped to three dif-ferent types of role.Then,this paper introduces the detailed flow chart of collab-orative computing,which can be flexibly applied to the crowd sensing systems.Based on the relative locations of the perceived users as well as the physical connectivity,the system dynamically assigns role type to the users and specifies the duration of the roles.Different roles work together to perform a perceptual task through collaborative computing.Furthermore,this paper formalizes the broadcaster set selection problem(BSSP)and proposes two heuristic algorithms to solve the BSSP.We use the real user trajectory data to verify the proposed collaborative computing method.The experimental results show that for the perceptual task that needs to obtain the location information,68%of the power can be saved through the collaborative computing.