Session-2 Participants Cyrus Shahabi Raju Vatsavai Mohamed Mokbel Shashi Shekhar Mubarak Shah Jans Aasman Monika Sester Wei Ding Phil Hwang Anthony Stefanidis (Matt Duckham) (Angelos Stavrou)
Making sense of spatiotemporal sensor data (GeoSensor Nexus) Collaborative Acquisition, Querying, and Spatiotemporal Analysis of distributed sensor data in geographic spaces
Defining Characteristics Cover all locations, all the time (with resource constraints) in 3D – Various resolution, quality/uncertainty, sparsity in space and time Support all types of sensors (Multi-modal, multi-source) – Remote sensors, moving sensors, humans as sensors, web as sensors, cameras, camcorders, cell-phones, microphones, RFID, human body sensors, chemical sensors Large scale of sensors (and hence data) Spatiotemporal analysis to identify, model and understand high- level events and processes; and then react to them – Timely spatiotemporal analysis of sensor data (present) – Archival and predictive spatiotemporal analysis of sensor data (past & future)
Sample Challenges Analysis On GPS data: noisy and uncertain data; clustering trajectories, classification (e.g., indoor/outdoor, walking/running) On Traffic sensors: bulky data; traffic prediction On remote sensors: not enough training data ; extracting associating rules On special-purpose sensors (e.g., radioactive detection): signature not unique, multi-feature; outlier detection On video/image sensors: non-structured data; track moving objects, geo-registering through video cameras Performance (e.g., response-time) On remote sensor and traffic sensor data: Multi-resolution aggregation and indexing Optimization On moving sensors (e.g., humans w/ cell phones, trucks): Planning for efficient and optimal acquisition of data at the right time and space
Sample Challenges … In-network processing – Add geo to sensor networks; adapt query processing techniques to restrictions of geo-sensors (e.g., power, bandwidth) – The local+distributed computation, e.g., improve the accuracy of GPS data by using multiple sensors (with differential GPS) – Distributed in-field computation of geo-sensors (e.g., monitoring propagation of a contaminant, tracking) – Track moving objects from a network of still and moving cameras Fusion of sensors (integration) – For forensic (identify when a video/image is taken, what was happening there, …) – Using geospatial sources to improve image/video analysis – With other traditional sources (different reliability and granularity) Privacy, Security & Trust – When people data are involved: Trajectory anonymization, Private LBS – Share sensor & resources without sharing data (network) – Dealing with adversarial data & un-trusted sources for analysis
A Different Perspective: Spatio-Temporal Scale (not necessarily in hierarchical order) Level 1: Process local raw data measured by a sensor, e.g. thresholding Level 2: Multi-sensor correlation for focal or teleconnection – Triangulation to position a moving object – Identify anomalies across sensors (e.g. discontinuity) Level 3: Aggregate common global operation picture – Extract events/processes, Interpret events in context, Develop hypothesis about current events – Knowledge Discovery - maps, descriptive models, visualization – Data Mining - Descriptive models: clusters, trends, associations,... Level 4: Prediction (e.g. via forward/inverse models) of process – Predict future states, e.g. final states (goals) and – Explain cause (intent/drivers/phase changes) Level 5 – Action, System Optimization: – How to redirect intelligence, surveillance, and reconnaissance (ISR) to improve performance (e.g. get better sensor utilization)
Backup Slides
Topics Raju: Sensor Net, ORNL government program – Data Processing: sensors measuring radioactive material (at truck weight station) – Challenges: signature not unique, outlier detection, multi-feature Wei: Remote sensing dataset: multispectral, analyze images to detect shapes (crater), integrating multi- layers (land-use, vegetation, temperatures) – Challenges: noisy data, too many false positives, not enough training data; complex structures Mubarak: image/video sensor; track moving objects from a network of still and moving cameras, fusion of videos; geo-registering through video cameras – challenges: using geospatial sources to improve image/video analysis Mohamed: add geo to sensor networks; adapt query processing techniques to restrictions of geo-sensors (e.g., power, bandwidth) Monika: specifics of sensor network is the local computation (local+distributed processing); where are the applications?; improve the accuracy of GPS data by using multiple sensors (with differential sensors) Jans: graph database (social networks): GPS sensors on moving objects and dealing with moving objects (historic and real-time decisions and planning) Tony: distributed in-field computation of geo-sensors (e.g., propagation of a contaminant, tracking) – Challenges: missing data, mobile sensors (actuate sensors); both real-time and historical, prediction. Human as sensors (text, speech), web sensors, surveillance (integrate multi-model and multi-source information to detect events) ; interacting/navigating the network Phil: sensors: traffic cameras, cell-phones; back-end: forensic (identified when is taken, what was happening there, …) Shashi: Analyzing/mining the measurements of sensors – several levels of processing: raw data analysis (denoising), simple aggregate queries (average, standard), data-mining (patterns), decision making (human analysis), power restriction is not an issue all the time
Topics …. Cyrus: Traffic sensors (mining; large size/response- time; compression) Humans as sensors (planning) GPS sensors (privacy LBS; mining: trajectory clusters, classification, e.g., indoor/outdoor, walking/running) Remote sensors (access to raw data at multiple resolutions, e.g., for visualization)