An Energy-Efficient Approach for Real-Time Tracking of Moving Objects in Multi-Level Sensor Networks Vincent S. Tseng, Eric H. C. Lu, & Kawuu W. Lin Institute of Computer Science and Information Engineering National Cheng Kung University Tainan, Taiwan, R. 0. C. Proceedings of the 11th IEEE International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA’05)
Outline Introduction Assumption Proposed Method Simulation Conclusions
Introduction (motive) In Object Tracking Sensor Networks Power consumption affects the lifetime capture a missing object back in real- time Energy efficiency and timeliness are the two important issues No work that considers both of real-time and energy efficient issues in OTSNs simultaneously
Introduction (solution) Propose a new approach Efficient and real-time tracking of the moving objects By mining the movement log Cluster the sensors Use the multi-level structure Predicting the next locations of moving objects in OTSNs First proposed a data mining method to discover the temporal movement patterns "Mining Temporal Movement patterns in Object Tracking Sensor Networks" First International Workshop on Ubiquitous Data Management, Tokyo, Japan, April, 2005
Assumption The sensors are distributed randomly The communication routing between sensors has been worked out The movement history of the moving object can be obtained from the OTSNs A server sensor in each sensor cluster server sensor can communicate with all the sensors within the region
Proposed Method Approach consists of three phases Clustering of sensor nodes Discovery of movement rules Prediction and recovery of moving objects definition Sequential path A sequence of sensors that were visited in time order by an object between its entering and leaving movement dataset The collection of movement paths generating from moving objects
Clustering of sensor nodes clustering mechanism K-means algorithm The goal is to divide the objects into K clusters K sensor nodes as initial centers Each node is assigned to its closest center The center of each cluster is re-calculated Until no change for the centers.
Clustering of sensor nodes Multi-Level Clustering of Sensor Nodes To construct the hierarchical structure Two import parameters Fun-out To model the branch of the hierarchical structure Height the depth of the hierarchical structure
Clustering of sensor nodes
Discovery of movement rules Mining of Movement Patterns Two kinds of movement patterns Sensor to sensor ex. Object moves from node a to node b Sensor to region ex. Object moves from node a to R11 The frequency of the inference rule Used to evaluate the confidence of the rule The highest frequent one serves as the basis of the prediction
Prediction and recovery of moving objects The movement rules To predict the next location for a moving object in the sensor networks Activate the least number of sensors Recovering To capture back the missing object Extend the scope of the region for sensor activation
An Illustrative Example Movement log of object Mining The movement rule
An Illustrative Example Level 0 represents the prediction of sensor- to-sensor Level 1 and Level 2 demonstrate the frequency of two levels Level 3 indicates the worst case that all sensors are activated
Simulation Average Search Time (AST) the average time required to recover the missing moving object Average Energy Consumption (AEC) the average energy consumption that is required to recover the missed moving object Miss Rate (MR) the rate that the search time required to recover the missing object exceeds the predefine deadline threshold
Simulation Impact of the number of sensor nodes
Simulation Impact of deadline threshold
Simulation Impact of the number of movement log
Simulation Impact of Fan-out and Height
Conclusions Proposed a prediction model based on multilevel architecture and clustering algorithms for tracking the objects in OTSNs Future work Consider multiple moving objects Consider many other factors Ex. representative of generated data