Approach - the WiSeNet Lab by liuhongmeiyes

VIEWS: 0 PAGES: 53

									CamNets: Coverage, Networking and Storage
 Problems in Multimedia Sensor Networks

                Nael B. Abu-Ghazaleh

      State University of New York at Binghamton and
             Carnegie Mellon University, Qatar
                 nael@cs.binghamton.edu



                           1
                     Talk Outline


• Introduction
• Overview of past work
• Current Active Research
  – Camera Networks
     • Camera coverage
     • Networking for data delivery and coordination
     • Storage and Indexing
• Future directions


                            2
Wireless Networks




Wireless Local Area Networks          Mesh networks




                    Sensor networks
             Sensor Networks
• What is a sensor network?
  – Sensing
  – Microsensors
  – Constraints, Problems, and Design Goals
Applications
                 Applications
• Interface between Physical and Digital Worlds
   – Many applications
• Military
   – Target tracking/Reconnaissance
   – Weather prediction for operational planning
   – Battlefield monitoring
• Industry: industrial monitoring, fault-detection…
• Civilian: traffic, medical…
• Scientific: eco-monitoring, seismic sensors, plume
  tracking…
       Microsensors for in-situ sensing



• Small
• Limited resources
   –   Battery powered
   –   Embedded processor, e.g., 8bit processor
   –   Memory: KB—MB range
   –   Radio: Kbps – Mbps, tens of meters
   –   Storage (none to a few Mbits)
                       Mica2 Mote

Chipcorn                 128KB Instruction
CC1000                                       UART       Flash
                            EEPROM
Radio Transciever                                      Memory
Max 38Kbps                                          128KB – 512KB
- Lossy transmission      4KB Data RAM


                          Atmega 128L
                          microprocessor
                          7.3827MHz

                                     UART, ADC

                          51 pin expansion
                             connector
                             Properties
• Wireless
   – Easy to deploy: ad hoc deployment
   – Most power-consuming: transmiting 1 bit ≈ executing 1000 instructions
• Distributed, multi-hop
   –   Closer to phenomena
   –   Improved opportunity for LOS
   –   radio signal is proportional to 1/r4
   –   Centralized apporach do not scale
   –   Spatial multiplexing
• Collaborative
   – Each sensor has a limited view in terms of location and sensor type
   – Sensors are battery powered
   – In-network processing to reduce power consumption and data
     redundancy
       Typical Scenario



     Deploy
                   Wake/Diagnosis




Self-Organize         Disseminate
Sensor Network Systems
Ghost of Research Past




          14
  Design Space and Infrastructure
            Tradeoffs
• We defined the design space for sensor
  networks
• Studied infrastructure and deployment
  alternatives
  – Identified congestion and its impact on sensor
    networks
     • New congestion management solutions
     • …including non-uniform information dissemination


                           15
                      Routing
• Real-Time Routing based on Just-in-Time-
  Scheduling
• Stateless Routing Protocols
  – Explain Anomalies in Virtual Coordinate Systems
  – Developed solutions that addressed them
     • Aligned Virtual Coordinates
     • Delivery guaranteed routing
     • Hybrid geographical/virtual routing protocols


                            16
        Sensor Network Storage
• Collaborative storage to reduce space and load
  balance

• Resolution Ordered Storage for space reclamation

• Interval summaries for indexing and coordination

• RESTORE testbed


                          17
       Localization and Security
• Securing Localization Systems

• Localization for Mobile Nodes: the self-tracking
  problem

• Trusted routing

• Defeating Timing and Space/Time Analysis attacks


                          18
Applications and Programmability
• Testbed for chemical/biological attack
  monitoring

• Camera Networks Testbed

• Filesystem abstraction for sensor networks

• Virtualizing sensor networks
                       19
Ghost of Research Present




           20
General Areas of Interest




      Modeling               Applications
     Simulation             Characterization
   Network testbed           Performance
   Robotic testbed             Security
Wireless Interference

 • Nodes interfere with each other

 • Effects
   • Lower throughput, Longer delays
   • Application performance

 • Our work
   • Understand and characterize interference
   • Design interference-mitigating protocols



                     A    B    C
    Example 1: Two-flow problems

• Only 2 links
• What are different ways in which they
  interact?
• How often do they occur?
• How does it affect throughput and delay?



A   B             A   B
        C   D             C   D     A   B D C
  Example 2: Application of interactions

Interaction Engineering
• Goal: Avoid harmful interactions
• Approach:
  – Detect interactions dynamically
  – Adapt parameters to overcome harmful
    interactions


     A   B   C   D                   A     B   C   D
Routing

• Transmit packets from source to destination
   o Link quality, scheduling and application-specific traffic.

• Our work: Study the optimal routing problem and heuristic
  protocols.
                      Congested!!
Example 2: Interference-aware routing

 Goal:
 Find routes that are aware
 of interference.


 Approaches:
   • Multi-objective optimization
   • Network-flow problems
   • Approximate heuristic
     protocols.
Testbeds

State-of-the-art wireless devices
 • Soekris boards, Software-Defined
   Radios

Current research projects:
 • Real-time models
    o Scheduling, routing
 • Efficient protocol development
    o Power control, rate-control, routing
 • Robotic projects
    o Camera-Nets
    o Localization
Example 3: Mesh Network Monitoring tool




Distributed measurement
protocols
 • Network Topology, Link
   Quality, ...
 • Detect interactions

Framework to build higher
level protocols.
                  Introduction

• A smart camera network is a network of autonomous
  and cooperative camera nodes.

• Traditional Camera Networks:




                        32
              Why are they interesting?


• Many applications
  – Military: sensitive areas
  – Homeland security: suspicious activities, aftermath
  – Disaster recovery: help rescue operations
  – Habitat monitoring: capture scientific information such as
    behavioral/migration patterns of animals
  – Road traffic monitoring: detect and report traffic violations




                              33
                   Motivation

• Problems with traditional networks:
  – Simple capture-and-stream nature:
     • needs human to monitor and control cameras.
  – Fixed and costly infrastructure:
     • high-end cameras, wired connectivity.


• An expectation from a smart camera network:
  – autonomously capture most useful information
    from the deployment region.

                        34
       Major Problems in Camera Networks
• Computer vision related problems
  – Camera calibration
  – Target detection and identification
  – Event classification and clustering

• Systems related problems
  –   Camera Coverage
  –   Network: Quality of Service for data delivery
  –   Network: Coordination
  –   Storage and indexing
                           35
       Coverage Maximization Problem
How to configure cameras’ FoVs to maximize the total
 number of targets covered?
   – Assuming all targets are equally important.


• Camera Configuration Parameters
   – Pan: horizontal adjustment
   – Tilt: vertical adjustment
   – Zoom: coverage range adjustment


• Camera Field-of-View (FoV):                      R
   – Represented by angle and depth of view

                              36
   Coverage Maximization Problem




– Assumptions
   • Discrete pans
   • Boolean coverage model
   • No occlusions

                              37
       Solution Approach


     Why not a greedy approach?




C1            C2        C3




               39
             Contributions


• Integer Linear Programming based formulation

• Centralized heuristic

• Semi-centralized approach for scalability




                     40
              ILP Formulation



Subject to:




                    41
   Centralized Approach for Solving ILP

• Each camera sends state information to a central node
   – State information: <Camera Id, Target Id, Target location>


• Central node computes optimal orientations (pans) for
  each camera and sends them back.

• The optimization problem is NP-hard!



                              42
 Centralized Force-directed Approach
                (CFA)
Approach: Iteratively choose camera-pan pair with
 highest force (Fik)

Example:
                                      M: set of
                                      targets
                                      N: set of
      F=1       F=0.5                 cameras
                                      P: set of
                                      pans

                 F=0.5



                         43
Centralized Force-directed Approach
               (CFA)
Algorithm:




                44
     Centralized Force-directed Approach
                    (CFA)
Counter Example:                     C2
                                     P2        P1
                                     C3
                           P2             P1        P2        P1

                                     P2        P1
                                                         C1
Force Matrix
 Camera        P1     P2                  C4
   C1          0.25   0.75
   C2          0.67   0.33
   C3          0.67   0.33
   C4          0.67   0.33

                                45
    Scalable Semi-centralized Approach
• Centralized approaches are not scalable
  – Exponential computations for optimal solution
  – Large response delay

• Hierarchical Approach
  – Address scalability by spatially decomposing
    camera nodes into multiple partitions.
  – Key Idea:
     • take advantage of physical separation among cameras,
       at a possible expense of coverage gain


                            46
        Spatial Partitioning Approach
• Single Linkage Approach (SLA)
  – Bottom-up clustering approach
  – Start by treating each camera as a cluster
  – Merge two clusters if the smallest distance (d) between any
    two nodes is smaller than threshold.
  – Keep increasing the threshold to merge more clusters,
    forming a hierarchy.
• Modifications in SLA:
  – Termination condition for merging: d > 2*Rsensing
  – Maximum cluster size (Smax)
                                             R          R


                             47
                Performance Evaluation

•       Simulations using QualNet network simulator
•       Parameters:
    –     FoV Rmax = 100 meters; Rmin = 0 meters
    –     FoV Angle = 45°
    –     Terrain 1000x1000 meters
•       Benchmarks:
    –     Centralized Greedy Approach (CGA) [Abouzeid’06]
    –     Distributed Greedy Approach (DGA) [Abouzeid’06]
    –     Pure Greedy Approach (Greedy)



                                48
  Study of varying number of targets
# Cameras = 50




     Random                            Clustered

Percent Coverage: Ratio of covered to coverable objects

                          49
 Study of varying number of cameras

# Targets = 100.




E2E delay: Worst-case delay to receive response.

                          50
       Scalable Coverage for Static Targets

Study of impact of Smax
#Cameras=50; #Targets=100; Terrain: 500x500m




                           51
             Coverage for Mobile Targets

• Problem:
  – How to maximize the total mobile targets tracked?


• Approach:
  – How to compute the camera configurations?
     • Optimal, CFA, Hierarchical
  – How often to compute the optimal solution?
     •   Locally: local collaboration approach
     •   Globally: periodic recalibration
     •   Collaboratively: on-demand recalibration
     •   Hybrids

                                  52
         Coverage for Mobile Targets

Comparison of different policies and their combinations




      Params: N = 20; Mobility: pedestrian mobility parameters

                              53
         Conclusion & Future Work
• Focused on the coverage maximization problem
• Proposed three solution approaches:
   – ILP based formulation
   – Centralized heuristic: CFA
   – Semi-centralized approach: Hierarchical
• Semi-centralized approach can reap benefits of
  centralized and distributed approaches
• Future Work:
   – Extend formulation for tilt and zoom
   – Model obstacles in the formulation
   – Propose approach for mobile targets case
                             54
Ghost of Research Future




           55
             Future Directions
• Immediate Future
  – Camera Networks
  – Software Defined Radios
  – Measurement based protocols
• Getting into
  – Cyber physical systems –Smart cities
  – Environmental Observatory Networks
     • Augmented with mobile sensing and personal sensing


                           56
                 Barrier Coverage
• Approach
  – Model the terrain as a Triangulated Irregular
    Network (TIN) [Goodchild95]




  – Model FoV by assuming each triangle as a planer
    tile
  – Choose minimum number of ‘connected’ triangles.
                         57
  Спасибо большое

какие-нибудь вопросы?



         58
                 Barrier Coverage
• Approach
  – Model the terrain as a Triangulated Irregular
    Network (TIN) [Goodchild95]




  – Model FoV by assuming each triangle as a planer
    tile
  – Choose minimum number of ‘connected’ triangles.
                         65

								
To top