Resource Constrained Target Tracking for Visual Sensor Networks

While the Particle Filter has been widely investigated for inference problems of nonlinear systems (target tracking, navigation, etc.), incorporating it in stand-alone systems is challenging. In our work, we investigated several instances of visual target tracking via Particle Filter and found that conventional approaches could not handle the computational complexity, energy, thermal, etc. demands. For instance, traditional schemes for tracking targets in video are receiver-based where images are obtained by a camera are transmitted to the receiver whose platform runs the Particle Filter. While this approach is appropriate when the camera has little or no computing capabilities, it requires considerable energy and bandwidth for transmission of the entire video stream. Alternative sender-based  architectures require the camera to perform all the target tracking with a Particle Filter and have their own issues as well. Sender-based approaches requires that the sender have the computational resources and available energy to execute complex computations and track the target. Neither receiver nor sender-based approaches are well-suited for cases where (i) sender and receiver both have computational resources; (ii) there is considerably more energy available at one side of the system compared to the other; or (iii) one side is experiencing a thermal emergency. 

The problem with the above approaches is that they end up overburdening the sender or receiver irregardless of present resources and current system conditions. In our work, we have proposed several distributed and adaptive solutions that were able to execute target tracking more effectively by dividing complex Particle Filtering steps (such as computing particle weights) between sender and receiver. Our approaches take advantage of two concepts: Workload Reduction and Workload Migration. Workload reduction is utilized to adapt resource usage as the behavior of the target being tracking changes. Workload migration uses an optimization framework to dynamically shift workload based on current availability of resources at sender and receiver. Simulation results showed that the proposed methods could maintain safe operating temperatures at the devices and lengthen target tracking lifetime compared to traditional architectures without substantial overheads. 

Our Conference and Journal Papers

NOTE: This directory contains pdf/ps files of articles that may be covered by copyright. You may browse the articles at your convenience, in the same spirit as you may read a journal or a proceedings article in a public library. Retrieving, copying, or distributing these files may violate copyright protection laws.

  • D. Forte, A.Srivastava, “Resource-Aware Architectures for Adaptive Particle Filter Based Visual Target Tracking”, ACM Transactions on Design Automation of Electronic Systems (TODAES), Vol. 18, No. 2, April 2013. [pdf]
  • D.Forte, A.Srivastava, “Adaptable Architectures for Distributed Visual Target Tracking”, Computer Design, 2011 IEEE International Conference on (ICCD) , Oct. 2011. [link]
  • D. Forte, A. Srivastava, “Resource-Aware Architectures for Particle Filter Based Visual Target Tracking”, Green Computing Conference and Workshops (IGCC), 2011 International, July 2011. [link]