Download presentation
Presentation is loading. Please wait.
Published byNickolas Booth Modified over 9 years ago
2
Tracking Mobile Sensor Nodes in Wildlife Francine Lalooses Hengky Susanto EE194-Professor Chang
3
Outline Recap Tracking Algorithms Failure and Recovery Algorithms Future Work References
4
Recap Purpose of tracking For wild life habitat watch purpose Sensors only monitor land targets (animals) Animals are tagged Sensors were purposely placed at certain location Save time finding target in the region Better understanding of region/animal relationship
5
Cow Tracking E-mail, 2004 Is it just me........ or does anyone else find it absolutely amazing that the U.S. government can track a cow born in Canada almost three years ago, right to the stall where she sleeps in the state of Washington, and determine exactly what that cow ate. They can also track her calves right to their stalls, and tell you what kind of feed they ate. But they are unable to locate 11 million illegal aliens wandering around in their country, including people that are trying to blow up important structures in the U.S. My solution is to give every illegal alien a cow as soon as they enter the country.
6
Tracking Algorithms Factors affecting tracking performance: Sensor range Average speed of target Algorithms: Naïve Activation Randomized Activation Selective Activation based on Prediction Low Duty Cycle Operation
7
Naïve Activation All nodes are in tracking mode all the time Worst energy efficiency Best possible quality of tracking Energy-Quality Tradeoffs for Target Tracking in Wireless Sensor Networks, USC.
8
Randomized Activation Each node is on with probability p
9
Selective Activation Based on Prediction Small subset of nodes are in tracking mode Nodes predict “next” position of target Rest of nodes in communication mode
10
Low Duty Cycle Operation: Frisbee Model Entire sensor network turns off and on Low power operating mode with wakeup Power-saving mode “Wakeup wavefront” Sensors must use localized algorithms Fully distributed, decentralized design Each node autonomously decides whether it lies in Frisbee Decentralized decision = which exact node should wake up
11
Frisbee Model
12
Failure and Recovery Recap Hierarchy cluster based sensor networks management Only Cluster Head (CH) communicates with other CH CH will wake up the next CH in target’s path Failure occurs when lost ACKs between cluster heads
13
Failure and Recovery Algorithms Failure and recovery factors: Broken link Power consumed CH dies Algorithms: Retry sending Space decomposition (hierarchical clustering) Sweeping across the region Search by region
14
Space Decomposition Quicker to find the lost target Takes O(log n) running time for a successful search Guaranteed to find lost target if the target is still in the region Awake every node in the region Not energy efficient Costly Creates network traffic
15
Sweeping Across the Region Sweeping outward from last seen position to border node Perform a short overlap layer search for fault tolerance Only notifies their neighbor at outer layer When successful, the founder takes over target When target is not found, border sensors report to base station Awake all nodes in region and flood network Running time is O(n) Sensor node layers Example of sweeping:
16
Search by Region Hierarchy cluster / tree environment CH and its subordinates Cluster border nodes CH makes decision based on input from its subordinates CH knows radius of its cluster Cluster size is proportional to other clusters Drawback: Constantly changing topology
17
Search by Region Approach Extending search to larger region Uses hypothesis or cluster border nodes Advantages: Minimizes the number of clusters involved Reduces network traffic Allows multilevel monitoring (hierarchy cluster based)
18
Cluster Border Nodes Summary Nodes at the edges of clusters Alternative approach to predicting the next target’s location Avoid fault prediction Helps CH’s decision of contacting the next proper CH All nodes ask their neighbor’s CH ID All nodes whose neighbors have different CH ID declare to be border node Border nodes report to its CH with new status and neighbor CH ID
19
Search by Region: Cluster Border Dark area: where the target is lost Algorithm: CH uses cluster border to determent the next location CH of target’s last position broadcast to all neighbors Only CHs attached to dark area wake up and continue broadcast Otherwise ignore the alert and sleep Drawback: Difficult to predict animal’s behavior without prior knowledge Difficult to determent if areas are covered properly How to determine if target is still in dark area? Proposed solution: Tagging target and retrieving clue from tag Use two E[X] to determine whether or not target is still in dark area
20
Search by Region: Hypothesis Take advantage of hierarchy cluster structure Each CH counts average visit per day by any target (e.g. animals) High rank CH queries clues from its subordinates Create a prediction based on hypothesis Find popular place to roam Predict a trace from predicted destination to last known location of animal
21
Search by Region: Foreseeable Issues How to bound the search area What is the probability of: Target electing to visit non-hypothesized destination Target taking different path to predicted destination Multiple candidates of popular destinations
22
Future Work Tracking algorithm Compare current tracking algorithms Implement better algorithm Failure and recovery algorithm Optimize current algorithm Solve problem at “ Search by Region: Foreseeable Issues” Propose new algorithm Simulation Performance analysis of the algorithms
23
References Energy-Quality Tradeoffs for Target Tracking in Wireless Sensor Networks. S. Pattem. USC, 2003. Habitat Monitoring: Application Driver for Wireless Communications Technology. D. Estrin, J. Zhao, et al. UCLA, 2000. Next Century Challenges: Scalable Coordination in Sensor Networks. D. Estrin, et al. USC, 1999. Sensing, Tracking, and Reasoning with Relations. L. Guibas. Stanford University, 2002. Computational Geometry. M. De Berg, M. van Kreveld, M. Overmars, O. Schwarzkopf. Utrecht University, 1999. Minimizing Communication Cost in Hierarchically Clustered Networks of Wireless Sensors. S. Bandyopadhyay, E.J. Coyle. Purdue University. Efficient Location Tracking Using Sensor Networks. H.T. Kung, D.Vlah. Harvard University. Distributed State Representation for Tracking Problems in Sensor Networks. J. Liu, M. Chu, J. Liu, J. Reich, F. Zhou. Microsoft Corp, 2004. Locating Moving Entities in Indoor Environments with Teams of Mobile Robots. M. Rosencrantz, G. Gordon, S. Thurn. Carnegie Mellon University, 2003. Lightweight Sensing and Communication Protocols for Target Enumeration and Aggregation. Q. Fang, F. Zhao, L. Guibas. Stanford University, 2003.
24
Questions
25
Backup Slides
26
Sweeping Across the Region Problem Running time is O(n) for a successful search The target might be able to fool the algorithm The target might leave the monitored area and return while search is performed and waste of searching effort The target might moves faster than the sweep because the network traffic might slowdown sweep High chance of flooding the network High probability of awake the entire sensors
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.