Presented at LTER ASM 03, 9/20/2003 Intelligent Sensor Network Signal and Information Processing For LTER Applications Yu Hen Hu University of Wisconsin – Madison Dept. Electrical and Computer Engineering Madison, WI 53706 hu@engr.wisc.edu © 2003 by Yu Hen Hu Outline • Long term environmental monitoring requirements • Intelligent signal and information processing (ISIP) – Intelligent agent and ISIP – Needs for intelligence • Enhance performance • Reduce cost • Work in progress – Detection of changes – Clustering of events – Sampling frequency determination • Future works – Intelligent monitoring – Intelligent maintenance – Intelligent data analysis 2 © 2003 by Yu Hen Hu Challenges in Long Term Environmental Monitor • Objectives and hypothesis – Specific biological, environmental phenomena to be observed and analyzed – Formulate hypothesis based on existing data • Signal and information processing – On-line, real time control • • • • Turn on/off sensor Adjust sampling frequency Data routing/collecting Streaming of video/image on-demand • Calibration and selfmonitoring • Experiment design – Location, duration of observation, – likelihood of onset of events of interests – Instrumentation • Types of sensors, • Deployment plan • Maintenance plan – Off-line • • • • data archival, retrieval Analysis, visualization inference – Data archiving plan 3 © 2003 by Yu Hen Hu Intelligent Agent – Intelligent agent are persistent software/hardware systems that perceive, reason, act, and communicate on behalf of human users Sensor Environment knowledge action Agent 4 © 2003 by Yu Hen Hu Characteristics of an IA • • • • • Autonomous execution Goal seaking Persistent within or as a part of a system Able to reason during action selection Acting for another with authority granted by another • Interact with other agents or human via dialog or some agent communication language 5 © 2003 by Yu Hen Hu ISIP: Intelligent Signal and Information Processing • Signal processing – The sampling, conditioning, compression, transmission, and analysis of numerical measurements of the environment based on sensor readings • Information processing – The handling of non-numerical data to coordinate collaboration, and control operations • ISIP – Perform signal and information tasks using intelligent agent – Tasks: • statistical and heuristic reasoning, including hypothesis testing, classification, estimation, data fusion, etc. – Tools: • neural network, expert system, fuzzy logic, genetic algorithm, pattern classifiers, time series analysis, statistical learning, support vector machine, Bayesian network, planning, etc. 6 © 2003 by Yu Hen Hu Wisconsin Long-Term Ecological Research (LTER) project Three Buoy Projects 1) Sparkling raft Serial communication Fixed buoy location Simple data format that must conform to historical format Wireless takes the place of serial cable, but download timing automated 2) 3 small roving buoys Serial communication Flexible buoy locations Complex data format with intensive post-processing Wireless simply takes the place of a serial cable 3) Large profiling buoy Bidirectional ethernet communication Fixed buoy location Simple but flexible data format Real-time data with web publishing and buoy control http://www.limnology.wisc.edu/ 7 © 2003 by Yu Hen Hu Wisconsin Limnology Lab Buoy Wireless Sensor Network 8 © 2003 by Yu Hen Hu How Buoy Data Are Processed? 9 © 2003 by Yu Hen Hu What a Buoy Will Do? 10 © 2003 by Yu Hen Hu Water Temperature Data temperature profile 30 water temperature 25 20 15 10 5 0 100 200 300 400 500 600 700 800 11 © 2003 by Yu Hen Hu A Change Detection Problem 12 © 2003 by Yu Hen Hu Sensor Node Signal Processing original time series and trend • Requirements: 22.5 – Simple algorithm – Robust performance – Low power operation 22 21.5 21 20.5 • Trend removal 100 – Use linear phase FIR filter 200 300 400 500 600 700 600 700 trend-removed time series 0.5 – Simple statistical method to detect outliers 7 • Outlier detection 0 -0.5 100 200 300 400 500 13 © 2003 by Yu Hen Hu 2 200 300 400 500 600 700 800 3 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 7 100 200 300 400 500 600 700 800 8 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800 15 14 13 12 11 10 6 5 4 100 9 1 5 0 -5 10 0 -1 10 0 -1 0.5 0 0 -0.5 10 0 -1 0.5 0 0 -0.5 10 0 -1 50 0 -5 50 0 -5 20 0 -2 20 0 -2 10 0 -1 10 0 -1 0.5 0 0 -0.5 0.5 0 0 -0.5 10 0 5 0 0 Decision Fusion 6 4 2 0 0 200 400 600 800 14 © 2003 by Yu Hen Hu Enabling Technologies • Enabling information technologies – – – – Internet networking and ad hoc mobile network Wireless communication Micro-electronic mechanical (MEM) devices System-on-chip technology integrating • Analog + digital • Sensing + wireless communication + processing + actuation 15 © 2003 by Yu Hen Hu Technical Challenges • Self-configuration, – – – – Self-organization Self maintenance Services discovery Directory services • Collaborative sensor signal processing – – – – – Sampling Encoding, compression aggregation Event detection Target identification, situation awareness – Tracking • Secure operation – Privacy protection • Ensure sensor information is accessed only by authorized personnel – Fault tolerance, high availability – Safety • Non-intrusive, • Safe actuation – Sabotage resistance, security 16 © 2003 by Yu Hen Hu Self-Configuration • Purpose – Reduce deployment and configuration cost • Vision – Sensors are deployed randomly (ad hoc network) to reach a desired local density – After deployment, sensors periodically communicate to each other to establish and maintain a connected network. – Directory (configuration information) will be aggregated, and published to authorized agent. – Sensors monitor network status and periodically report to an external monitoring agent – Sensor network re-organize itself in case • Its mission is changed as directed from an authorized external agent • Traffic/load changed over time • Sensors’ physical position changed if they are mobile • Part of the network malfunction due to sensor failure, or communication failure 17 © 2003 by Yu Hen Hu Self Configuration • How to join a physical network; that is, how is it authorized and given a network address and a network identity? • Once an entity is on the network and wishes to provide a service to other entities on the network, how does it indicate that willingness? • If an entity is looking for a service on the network, how does it go about finding that service? • How does geographic location affect the services an entity can discover or select for use? 18 © 2003 by Yu Hen Hu Collaboration • Sensors collaborate to achieve network-wide processing objective: – Higher performance – Lower resource (energy, band-width) consumption • Challenge: How to achieve globally optimal results – Using local, distributed criteria – With local communication – Using minimum amount of energy • Approach – Exploring redundancy and correlation • Densely deployed sensor field • Sensor readings are correlated 19 © 2003 by Yu Hen Hu Collaborative Sensor Signal Processing • Sampling – How to use lowest sampling rate/density to achieve desired spatialtemporal accuracy? • Compression – How to reduce overall sampled data that needs to be transmitted over wireless channel? • Aggregation – How to summarize information from multiple sensor without overload wireless channel • Event detection – How to deploy sensors so that a desired event can be detected with a specified accuracy? • Target identification, situation awareness – How to classify targets when there are multiple targets present • Tracking – How to coordinate sensor activities to track multiple target effectively. 20 © 2003 by Yu Hen Hu Sampling • Temporal resolution – How many samples per unit time? – Reduce rate to conserve energy, band-width – Based on underlying physics – Nyquist theorem for bandlimited signals – Adaptive sampling rate for non-stationary signals • Spatial resolution – For visual signals – How small the frame size can still allow • • • • Target detection Subject identification Tracking Other monitoring functions – Adaptive spatial resolution • Higher resolution in region of interests • Spatial-temporal sampling – Different camera/sensors coordinate to sample at lower rate while achieving higher resolution than a single camera 21 © 2003 by Yu Hen Hu Compression • • • • Sensors closer to each other physically, may sample similar (correlated) data It waste energy and bandwidth to transmit data uncompressed. Exploiting the correlation of sensor data among adjacent sensors, amount of data transmitted can be further reduced. Example: compression of multiple video streams taken from neighboring cameras • • • • • If sensor A reading = x, then sensor B reading = x 2, and vice versa Suppose both readings have range [0, 127] If reading A = 25, then sensor A knows sensor B’s reading in [23, 27]. If sensor B sends its reading in 7 bits, sensor A knows the receiving end must know reading A in [21, 29] Needs only 3-bits to encode! D. Slepian, and J. K. Wolf, “Noiseless coding of correlated information sources,” IEEE Trans. Information Theory, vol. 19, 1973, pp. 471-480 22 © 2003 by Yu Hen Hu Aggregation • Problem Statement – Find an estimate of a statistic of sensor measurements of a group of sensors with minimum amount of wireless transmission • Assume – Sensors can overhear neighboring sensor’s transmission – Sensor readings are correlated • Example: – Find maximum reading among N sensors • One possible protocol: – Sensor with higher reading report first. – Sensors with readings larger than reported readings will report with collision control – Sensors whose readings smaller than reported readings remain silent. – Wait for a pre-specified time period. The last reported reading is the maximum with high probability 23 © 2003 by Yu Hen Hu Event/Target Detection • Statistical hypothesis testing tasks • Collaborative detection – Correctly detect an event or a target while minimizing cost (energy and band-width) – Individual sensors may not detect correctly due to • Limited range, • Limited scope • noise • Methods of collaborative detection – Decision fusion – Multi-modality detection • Low cost sensor detect first with higher false alarm rate • High cost sensor (visual sensors) to verify detection results • Challenges – Not all sensors report detection – Not all reported detection will be forwarded to fusion center 24 © 2003 by Yu Hen Hu Target classification/ Event awareness • Pattern classification problems – Low power feature extraction – Decision fusion • Feature extraction – Invariant to variations – Cheap to compute – Local to each sensor • Decision fusion – Only discrete set of decisions needs to be transmitted over wireless channel • Event awareness – Detecting a particular event such as traffic accident – Require understanding of a sequence of states using hidden Markov model – Requires detection of onset and offset of an event – Requires tracking of objects of interests 25 © 2003 by Yu Hen Hu Conclusion • Sensor network is a new application area for computer vision, graphics and image processing • It requires multi-modality, multimedia processing under the constraint of minimizing communication and energy consumption. 26