International Journal of Parallel Programming

, Volume 42, Issue 4, pp 663–680

Engineering Energy Efficient Visual Sensor Network Applications Using Skeletons

  • Stefano Chessa
  • Susanna Pelagatti
  • Nicoletta Triolo
Article

DOI: 10.1007/s10766-013-0260-y

Cite this article as:
Chessa, S., Pelagatti, S. & Triolo, N. Int J Parallel Prog (2014) 42: 663. doi:10.1007/s10766-013-0260-y

Abstract

Visual sensor networks (VSNs) perform complex scene analysis algorithms that require significant computations and communications. Under this respect, the use of skeletons contributes to reduce the complexity of VSN programming and may ensure an easier and better optimization of the code. In this context, we propose INS, a stencil based skeleton targeted for wireless/visual sensor networks (W/VSNs) and give a preliminary analysis of its benefits using tracking as a case study. INS abstracts a distributed approximation schema in which the estimation of a given metric is organized in a sequence of steps. Each step includes collecting estimates from some neighbor nodes and local computation of a new approximation. In particular, INS takes inspiration from some stencil based skeletons proposed for parallel computation and merges it with the classical event driven model typical of sensor programming. As a result, the execution of each step is triggered by the detection of a relevant event in the environment. Tracking consists in periodically predicting position and velocity of one or more mobile targets. We discuss how INS can be instantiated to a distributed version of Kalman filtering. As energy efficiency is central in W/VSNs, we derive analytic models for energy dissipation of the INS skeleton depending on different concepts of neighborhood for the data exchanged at each step. Then, these models are used to guide the deployment of our tracking application on a real scenario.

Keywords

Visual sensor networks Algorithmic skeletons Energy saving Kalman filtering 

Copyright information

© Springer Science+Business Media New York 2013

Authors and Affiliations

  • Stefano Chessa
    • 1
  • Susanna Pelagatti
    • 1
  • Nicoletta Triolo
    • 1
  1. 1.Dipartimento di InformaticaUniversità di PisaPisaItaly