On statistical approaches to target silhouette classification in difficult conditions

https://doi.org/10.1016/j.dsp.2007.05.008Get rights and content

Abstract

In this paper we present a methodical evaluation of the performance of a new and two traditional approaches to automatic target recognition (ATR) based on silhouette representation of objects. Performance is evaluated under the simulated conditions of imperfect localization by a region of interest (ROI) algorithm (resulting in clipping and scale changes) as well as occlusions by other silhouettes, noise and out-of-plane rotations. The two traditional approaches are holistic in nature and are based on moment invariants and principal component analysis (PCA), while the proposed approach is based on local features (object parts) and is comprised of a block-by-block 2D Hadamard transform (HT) coupled with a Gaussian mixture model (GMM) classifier. Experiments show that the proposed approach has good robustness to clipping and, to a lesser extent, robustness to scale changes. The moment invariants based approach achieves poor performance in advantageous conditions and is easily affected by clipping and occlusions. The PCA based approach is highly affected by scale changes and clipping, while being relatively robust to occlusions and noise. Furthermore, we show that the performance of a silhouette recognition system subject to mismatches between training and test angles of silhouettes (caused by an out-of-plane rotation) can be considerably improved by extending the training set using only a few angles which are widely spaced apart. The improvement comes without affecting the performance at “side-on” views.

Section snippets

Dr. Conrad Sanderson is a researcher at NICTA and an adjunct research fellow at the Australian National University. He received the Ph.D. degree in 2003 from Griffith University (Queensland, Australia), in the area of information fusion applied to speech- and face-based biometric person recognition. He has worked on a number of applied research projects, including robust speech recognition at the Advanced Telecommunication Research Laboratories (Japan), audio-visual biometrics at the IDIAP

References (42)

  • D. Reynolds et al.

    Speaker verification using adapted Gaussian mixture models

    Digital Signal Process.

    (2000)
  • G.R. Doddington et al.

    The NIST speaker recognition evaluation—Overview, methodology, systems, results, perspective

    Speech Commun.

    (2000)
  • N. Nandhakumar et al.

    Robust thermophysics-based interpretation of radiometrically uncalibrated IR images for ATR and site change detection

    IEEE Trans. Image Process.

    (1997)

  • Chambers Science and Technology Dictionary

    (1991)
  • J. Ratches et al.

    Aided and automatic target recognition based upon sensory inputs from image forming systems

    IEEE Trans. Pattern Anal. Machine Intell.

    (1997)
  • J. Mitzel

    Multitarget tracking applied to automatic target recognition with an imaging infrared sensor

  • P. Bharadwaj, P. Runkle, L. Carin, Infrared-image classification using expansion matching filters and hidden Markov...
  • S. Der et al.

    Probe-based automatic target recognition in infrared imagery

    IEEE Trans. Image Process.

    (1997)
  • J.A. Alves, Recognition of ship types from an infrared image using moment invariants and neural networks, MS thesis, US...
  • V. Gouaillier, L. Gagnon, Ship silhouette recognition using principal components analysis, in: Applications of Digital...
  • D.N. Kato, R.D. Holben, A.S. Politopoulos, B.H. Yin, Ship classification and aimpoint maintenance, in: Infrared Systems...

    • Cited by (9)

    • Template matching of occluded object under low PSNR

      2013, Digital Signal Processing: A Review Journal

    • Detection and tracking of ships in open sea with rapidly moving buoy-mounted camera system

      2012, Ocean Engineering
      Citation Excerpt :

      Third, because of the importance of horizon detection in our algorithm, in Section 2.3, we briefly review the existing approaches to horizon line detection. The harbor, port, and coast surveillance have been extensively studied for more than a decade (Voles et al., 1999; Socek et al., 2005; Seibert et al., 2006; Feineigle et al., 2007; Rodriguez Sullivan and Shah, 2008; Chen, 2008; Sanderson et al., 2008; Bloisi et al., 2009). In general, all the existing systems for port surveillance assume that the camera is in a fixed position, and the water surface is fixed in terms of a camera view.

    • Key features for ATA/ATR database design in missile systems

      2017, Proceedings of SPIE - The International Society for Optical Engineering

    • Communication-constrained multi-AUV cooperative SLAM

      2015, Proceedings - IEEE International Conference on Robotics and Automation

    • Context modeling combined with motion analysis for moving ship detection in port surveillance

      2013, Journal of Electronic Imaging

    • Bridging the fuzzy, neural and evolutionary paradigms for automatic target recognition

      2011, Applied Intelligence
    View all citing articles on Scopus

    Dr. Conrad Sanderson is a researcher at NICTA and an adjunct research fellow at the Australian National University. He received the Ph.D. degree in 2003 from Griffith University (Queensland, Australia), in the area of information fusion applied to speech- and face-based biometric person recognition. He has worked on a number of applied research projects, including robust speech recognition at the Advanced Telecommunication Research Laboratories (Japan), audio-visual biometrics at the IDIAP Research Institute (Switzerland), ship classification in infra-red images at the Centre for Sensor Signal and Information Processing (CSSIP), as well as natural language processing (author identity deduction) and bioinformatics (feature selection for cancer classification) at NICTA. His current research interests are in the areas of machine learning, pattern recognition and computer vision, with applications such as intelligent surveillance.

    Dr. Danny Gibbins is a senior research fellow with the Sensor Signal Processing Group at the University of Adelaide. He received the Ph.D. degree in machine vision from Flinders University in South Australia in 1995. Between 1995 and 2005 he was a research fellow with the CRC for Sensor Signal and Information Processing (CSSIP) where he was involved in the development of surveillance, classification and recognition systems for radar and electro-optical systems. Since 2005 he has focussed on video surveillance and 3D terrain modelling applications for unmanned air vehicles (UAVs). His research areas include signal and image processing, object classification and automatic target recognition and 3D reasoning using LADAR.

    Stephen Searle completed the B.Sc. (Hons) at Flinders University of South Australia in 1991 and the M.App.Sc. at the University of Adelaide in 1997. His thesis entitled “Matched Doppler Processing” examined the modelling of Doppler effect in matched field processing solutions for moving target localisation with acoustic measurements. Since 1992 he has worked for government, academia and industry in the broad area of digital signal processing research. Applications have included sonar signal processing and array shape estimation, communications signal processing (in particular baseband CDMA simulation), motion detection and tracking with visual-band and infrared imagery, target detection and parameter estimation. Since 2006 Stephen has been employed as a research fellow by the University of Melbourne. His current research interests include target motion analysis, radar signal processing and waveform design.

    View full text
    Copyright © 2007 Elsevier Inc. All rights reserved.