Project topics and materials
Get Complete Project Material File(s) Now! »
Mobile Crowdsourcing Platforms
Mobile crowdsourcing has gained a lot of attention for the last decade [13, 73], and this is due to the capabilities that offer these mobile phones. Scientists and decision makers can now run their experiments on the user mobile device to collect real life scenarios. The collected data can be of different types: geospatial data [13], behavioral data [74], environment monitoring [75, 76], Internet quality monitoring [77] and health data [78].
To help scientists, crowdsourcing platforms leverage all the technical kit to deploy their crowdsourcing collection campaign, to monitor the campaign and to store the collected data in the cloud servers. In our work, we test our privacy preserving techniques on both APISENSE crowdsourcing platform and MobiPerf crowdsourcing mobile app: The crowdsourcing platform APISENSE [79] offers the scientists, with no technical back-ground, the ability to acquire insightful data from the field. APISENSE manages the deployment of the dada acquisition campaigns as dedicated tasks described in JavaScript, which are remotely deployed to all the participants in real-time. APISENSE offers a generic mobile app that can be used by various scientists for different use cases. This mobile app is in charge of directly executing the deployed tasks on the participant’s mobile device. The collected data comes from most of the sensors available on a mobile device, but the scientists can also ask the participant to perform a specific action whenever a given event happens, like answering a question whenever the participant comes back home. The APISENSE platform provides a web interface to monitor and to control the crowd-sourcing campaign. Through this web interface, the campaign manager (the scientist) can invite new users, update the data acquisition script and update it in real-time. The campaign manager can visualize the anonymized user data, as soon as they are uploaded to the server. Although, these received data do not contain any user identifiers, it needs to be further anonymized before starting to process the collected data. MobiPerf is open source mobile app [27] for measuring network performances at regular intervals in the background. The MobiPerf mobile app use the open source library Mobi-lyzer [80] that facilitates the network measurement crowdsourcing campaigns. It contains most of the network measurement tools which can be used just by configuring a file that contains the measurement tasks (e.g., ping, traceroute, HTTP GET) to be executed on the user mobile device. To use the MobiPerf mobile app, the scientists need to customize the app to meet their needs and to change the server address by their own server address.
WiFi Places Collection
Mobile devices are equipped with a variety of sensors. Therefore, they can collect insightful data about both the user surrounding environment and the user activities to feed the mobile apps with the user context. These context data include the user GPS location and the surrounding WiFi networks.
One of the methods to contextualize the user with WiFi networks is the indoor positioning systems and WiFi fingerprinting systems. These systems use the surrounding scanned WiFi Access Points (APs) to localize indoor users or to track users’ places.
WiFi Indoor Positioning Systems
Due to the absence of the Global Positioning System (GPS) signal inside buildings, many new systems have been proposed as an indoor positioning system [87]. Among these systems, WiFi-based systems, which take advantage of the already spread WiFi Access Points (APs) and the users phone that are equipped with WiFi scanning abilities. WiFi signals can be used to calculate the position of a mobile device down to 1.5 meters precision [88]. While some research works require a prior training phase (online mode) to survey the location of the antennas in the map [89, 90], other proposals have been proposing an automatic way to build a radio map without any prior knowledge about the antennas [91–93]. In particular, Liu et al. [94] propose a peer assisted localization method to improve localization accuracy. Gorski et al. [95] present a multi-building indoor localization system using WiFi fingerprint based on the K nearest neighbors (KNN) method. Salazar et al. [96] use a Type-2 fuzzy inference systems to cluster WiFi fingerprints to localize the indoor users. Li et al. [97] introduce a privacy preserving WiFi indoor localization system. They argue that the localization query can inevitably leak the client location and lead to potential privacy violations. Jin et al. [98] propose a real-time WiFi positioning algorithm with the assist of inertial measurement unit to overcome the RSS variation problem. Pulkkinen et al. [89] present an automatic fingerprint population using theoretical properties of radio signals. They rely on the locations of WiFi APs and collecting training measurements. Salamah et al. [99] use the Principle Component Analysis (PCA) to reduce the computation cost of the WiFi indoor localization systems based on machine learning approach. Yiu et al. [100] apply training measurements and a combined likelihood function from multiple APs to measure the indoor and the outdoor user position. Capurso et al. [101] present an indoor and outdoor detection mechanism which can be used to optimize GPS energy usage. Yiu et al. [90] review the various methods to create the radiomap. Then, they examined the different aspects of localization performance like the density of WiFi APs and the impact of an outdated radiomap. Ahmed et al. [102] provide a new optimized algorithm for fast indoor localization using WiFi channel state information. Caso et al. [103] introduce an indoor positioning system that relies on the Received Signal Strength (RSS) to generate a discrete RSS radiomap. Li et al. [93] present SoiCP, a seamless outdoor-indoor crowdsourcing positioning system without requiring site surveying. Crowdsourced WiFi signals are used to build a radiomap without any prior knowledge.
WiFi Fingerprinting Systems
Due to the cost-effective of WiFi signals [104] against GPS regarding battery usage and accuracy, a lot of recent work in user contextualization were interested in using WiFi signals to contextualize the users both indoor and outdoor. With the location permission granted, surrounding WiFi APs can be provided to the app, these WiFi APs will act as the place fingerprint, different WiFi APs signals received mean different places. Tracking the user for several hours using just WiFi scans can provide a valuable information about the user activities [105], personality traits [106] and the user routines [107]. These data can also be used to find a unique fingerprint per users [108, 109], find social networks [110] or track users in an offline mode [111, 112]. To localize the user, her WiFi fingerprint has to be compared to other collected WiFi fingerprints. Once a similar fingerprint is found, the corresponding location is attributed to the user. Some research works are focusing on optimizing the similarity distance between two WiFi fingerprints [113, 114, 100, 115]. In particular, Zhang et al. [116] introduce POLARIS, a localization system using cluster based solution for WiFi fingerprints. They explain how to collect and compress a city scale fingerprints, then, how to find the location of a user using similar WiFi fingerprints. Sakib et al. [117] present a method to contextualize the mobile user using the WiFi APs fingerprints, with a simple clustering algorithm that creates a place for each group of WiFi APs. They validate their results with cellular cell ids datasets. Sapiezynski et al. [118] use WiFi to enhance GPS traces localization, with one GPS location per day; they can localize 80% of mobility across the population. Wind et al. [119] use WiFi scans to infer stop locations for users, their algorithm can tell if a user is moving or in a stationary state using just WiFi fingerprints. Choi et al. [120] propose a method for an energy efficient WiFi scanning for user contextualization; they try to minimize the number of scans depending on the scanned WiFi APs.
Synthesis
While a lot of work have been done to localize the user using WiFi indoor localization techniques, most of them require either a training phase or user input to locate the user using the WiFi antennas. Several research works have been focusing on the WiFi fingerprinting based on the clustering algorithms and similarity distances, but they do not report on the utility of the collected data, its cost or its privacy preserving methods. Furthermore, they do not provide any libraries or frameworks in order for their methods to be used and further improved. The WiFi places collection methods and algorithms have to be well documented and the resulted data have to be analyzed so that new contributions can be built on top of it.
Testing P2P Mobile Apps
Most of the reviewed privacy preserving methods propose to collaborate between mobile apps’ users using peer-to-peer communications to hide the workers from the crowdsourcing server. In this section, we overview the state of the art approaches used to test these peer-to-peer mobile apps. To overview these testing approaches, we first start overviewing the works that have been done in testing the mobile apps in general.
Table of contents :
List of figures
List of tables
1 Introduction
1.1 Motivation
1.2 Problem Statement
1.3 Thesis Goals
1.4 Contributions
1.5 Publications
1.5.1 Publication details
1.5.2 Proof of Concept
1.5.3 Awards
1.6 Thesis Outline
2 State of the Art
2.1 Anonymous Data Collection
2.1.1 Location Privacy
2.1.2 Mobile Crowdsourcing Platforms
2.1.3 Synthesis
2.2 WiFi Places Collection
2.2.1 WiFi Indoor Positioning Systems
2.2.2 Synthesis
2.2.3 WiFi Fingerprinting Systems
2.2.4 Synthesis
2.3 Testing P2P Mobile Apps
2.3.1 Testing Mobile Apps
2.3.2 Testing Peer-to-Peer Systems
2.3.3 Testing Nearby P2P Mobile Apps
2.3.4 Synthesis
2.4 Conclusion
3 Testing Nearby Peer-to-Peer Mobile Apps at Large
3.1 Introduction
3.2 Background & Motivations
3.2.1 Overview of the Google Nearby Framework
3.2.2 Presence of Google Nearby in the Play Store
3.3 Testing Nearby P2P mobile apps
3.3.1 Objectives when Testing Nearby Mobile Apps
3.3.2 Challenges for Testing Nearby Apps
3.4 The ANDROFLEET Testing Framework
3.4.1 Describing Test Scenarios in ANDROFLEET
3.4.2 Peer Discovery in ANDROFLEET
3.4.3 Running Test Scenarios in ANDROFLEET
3.4.4 ANDROFLEET Implementation Details
3.4.5 Case Study
3.4.6 Discussion
3.5 The PEERFLEET Testing Framework
3.5.1 Framework Overview
3.5.2 The PEERFLEET Orchestrator
3.6 Evaluation
3.6.1 Proximity Datasets
3.6.2 Bug Identification
3.6.3 Parameter Tuning
3.6.4 Threats to Validity
3.7 Conclusion
4 FOUGERE: User-Centric Location Privacy in Mobile Crowdsourcing Apps
4.1 Introduction
4.2 Privacy Threats in Mobile Crowdsourcing Systems
4.3 FOURGERE: Empowering Workers with LPPMs
4.4 Implementation Details on Android
4.4.1 Privacy Settings
4.5 Evaluations of FOUGERE
4.5.1 Evaluation Protocol
4.5.2 Empirical Evaluation
4.6 Threats to Validity
4.7 Conclusion
5 Privacy preserving WiFi Places for Mobile User Contextualization
5.1 Introduction
5.2 Motivation
5.3 User WiFi Place
5.3.1 WiFi place identification
5.3.2 User Place Building
5.4 Shared WiFi Places
5.4.1 Building atop of the WiFi Places
5.4.2 Extending the WiFi Places
5.5 WiFi Places Overview
5.5.1 WiFi Places Recognition Algorithm
5.5.2 The similarity function
5.5.3 Place notification
5.5.4 Place timetable
5.6 Data Collection
5.7 Results
5.7.1 Privacy Attacks
5.7.2 WiFi scans analysis
5.7.3 Evaluation metrics
5.7.4 Shared places
5.7.5 WiFi scan size
5.7.6 Place expansion
5.7.7 Discussion
5.8 Conclusion
6 Conclusions and Perspective
6.1 Contributions
6.2 Perspectives
6.2.1 Short-term Perspectives
6.2.2 Long-term Perspectives
References
