Home // SIGNAL 2017, The Second International Conference on Advances in Signal, Image and Video Processing // View article

Robust Object Tracking Using Unreliable Object Recognizers

Authors:
Li Li
Masood Mortazavi

Keywords: video surveillance; object tracking; camera network; baysian network; hidden markov model; forward-backward algorithm; Viterbi algorithm

Abstract:
This paper presents a method for video surveillance systems to correct noisy observations from distributed object recognizers that are unreliable. An unreliable recognizer consists of a hardware sensor (e.g., a camera) and a recognition program (e.g., facial recognition) that may produce random errors including false positives, false negatives, or failures. To address this issue, we use a Bayesian Network (BN) to connect multiple factors that can cause the noisy observations with random errors. We then incorporate the BN into an extended Hidden Markov Model (HMM) to infer optimal object paths from noisy observations. A prototype system is implemented and the simulation tests show that the Forward-Backward algorithm can achieve 77.3% accuracy on average with 47.9% relative improvement over the Viterbi algorithm. Both algorithms are robust to increasing noises and errors in the observations, even when 100% observations have over 66% errors.

Pages: 21 to 26

Copyright: Copyright (c) IARIA, 2017

Publication date: May 21, 2017

Published in: conference

ISSN: 2519-8432

ISBN: 978-1-61208-559-3

Location: Barcelona, Spain

Dates: from May 21, 2017 to May 25, 2017