Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Performance Assessment of Feature Detection Methods for 2-D FS Sonar Imagery

Hitesh Kyatham, Shahriar Negahdaripour, Michael Xu, Xiaomin Lin, Miao Yu, Yiannis Aloimonos

TL;DR

The paper addresses robust feature detection for 2-D forward-look sonar imagery in challenging underwater environments by evaluating eight detectors (SIFT, SURF, FAST, ORB, BRISK, SU-BRISK, F-SIFT, KAZE) on real data from five FL sonar devices. It uses two datasets collected in controlled aquatic facilities and employs Oculus as a reference to quantify cross-sensor feature detection via affine alignment and distance thresholds, highlighting detector sensitivity to device FoV and lens distortions. The study reveals detector-specific performance patterns across sensors, with SURF and FAST often yielding the highest feature counts while some detectors underperform on smaller features or certain devices. These findings guide detector selection for sonar-based localization, mapping, and navigation and point to future work on expanding real-world sonar datasets and developing adaptive, device-aware feature detectors for robust underwater perception.

Abstract

Underwater robot perception is crucial in scientific subsea exploration and commercial operations. The key challenges include non-uniform lighting and poor visibility in turbid environments. High-frequency forward-look sonar cameras address these issues, by providing high-resolution imagery at maximum range of tens of meters, despite complexities posed by high degree of speckle noise, and lack of color and texture. In particular, robust feature detection is an essential initial step for automated object recognition, localization, navigation, and 3-D mapping. Various local feature detectors developed for RGB images are not well-suited for sonar data. To assess their performances, we evaluate a number of feature detectors using real sonar images from five different sonar devices. Performance metrics such as detection accuracy, false positives, and robustness to variations in target characteristics and sonar devices are applied to analyze the experimental results. The study would provide a deeper insight into the bottlenecks of feature detection for sonar data, and developing more effective methods

Performance Assessment of Feature Detection Methods for 2-D FS Sonar Imagery

TL;DR

The paper addresses robust feature detection for 2-D forward-look sonar imagery in challenging underwater environments by evaluating eight detectors (SIFT, SURF, FAST, ORB, BRISK, SU-BRISK, F-SIFT, KAZE) on real data from five FL sonar devices. It uses two datasets collected in controlled aquatic facilities and employs Oculus as a reference to quantify cross-sensor feature detection via affine alignment and distance thresholds, highlighting detector sensitivity to device FoV and lens distortions. The study reveals detector-specific performance patterns across sensors, with SURF and FAST often yielding the highest feature counts while some detectors underperform on smaller features or certain devices. These findings guide detector selection for sonar-based localization, mapping, and navigation and point to future work on expanding real-world sonar datasets and developing adaptive, device-aware feature detectors for robust underwater perception.

Abstract

Underwater robot perception is crucial in scientific subsea exploration and commercial operations. The key challenges include non-uniform lighting and poor visibility in turbid environments. High-frequency forward-look sonar cameras address these issues, by providing high-resolution imagery at maximum range of tens of meters, despite complexities posed by high degree of speckle noise, and lack of color and texture. In particular, robust feature detection is an essential initial step for automated object recognition, localization, navigation, and 3-D mapping. Various local feature detectors developed for RGB images are not well-suited for sonar data. To assess their performances, we evaluate a number of feature detectors using real sonar images from five different sonar devices. Performance metrics such as detection accuracy, false positives, and robustness to variations in target characteristics and sonar devices are applied to analyze the experimental results. The study would provide a deeper insight into the bottlenecks of feature detection for sonar data, and developing more effective methods
Paper Structure (14 sections, 8 figures)

This paper contains 14 sections, 8 figures.

Table of Contents

  1. sec:intro
  2. Related Work
  3. Experiment and Analysis
  4. Setup and Dataset
  5. Assessment Methodology
  6. Pre-Processing
  7. Results
  8. First Dataset
  9. Second Dataset
  10. Conclusions
  11. Feature Count Analysis
  12. Detector Performance
  13. Common Feature Count Analysis
  14. Final Comments

Figures (8)

  • Figure 1: Overview of the setup and sample sonar images: (a) Experimental setup, (b) Oculus, (c) Gemini, (d) BlueView, (e) Didson, (f) Aris.
  • Figure 2: Lens distortion in lens-based Didson and Aris sonar has to be rectified to satisfy ideal sonar projection model (a) distorted Aris image (red channel) does not aligned correctly with overlapping Oculus image (green channel). (b) two image align correctly after distortion correction with a mild cubic model.
  • Figure 3: First data set comparative analysis of feature detection performance among eight detectors (SIFT, SURF, FAST, ORB, BRISK, SU-BRISK, F-SIFT, and KAZE) using four sonars (Oculus, Aris, Didson, and Gemini). Bar graph illustrates the number of features detected by each sonar-detector combination for a fixed sonar position. Each of the five subplots, represents a distinct sonar position.
  • Figure 4: First dataset - assessing effectiveness of different detector in identifying common features in images of different sonar types. The y-axis in each graph represents the number of common features in sonar pairs, showing comparisons of Oculus vs. Aris (purple line), Oculus vs. Didson (green line), and Oculus vs. Gemini (magenta line). The x-axis represents the distance threshold [pix]. Each of the 8 columns in the array corresponds to a different position.
  • Figure 5: First dataset - average number of features detected by various sonar systems (Oculus, Aris, Didson and Gemini) for different feature detection algorithms.
  • ...and 3 more figures