Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Entropy-Driven Sensor Selection in a Camera-LiDAR Particle Filter for Single-Vessel Tracking

Andrei Starodubov, Yaqub Aris Prabowo, Andreas Hadjipieris, Ioannis Kyriakides, Roberto Galeazzi

Abstract

Robust single-vessel tracking from fixed coastal platforms is hindered by modality-specific degradations: cameras suffer from illumination and visual clutter, while LiDAR performance drops with range and intermittent returns. We present a heterogeneous multi-sensor fusion particle-filter tracker that incorporates an information-gain (entropy-reduction) adaptive sensing policy to select the most informative configuration at each fusion time bin. The approach is validated in a real maritime deployment at the CMMI Smart Marina Testbed (Ayia Napa Marina, Cyprus), using a shore-mounted 3D LiDAR and an elevated fixed camera to track a rigid inflatable boat with onboard GNSS ground truth. We compare LiDAR-only, camera-only, all-sensors, and adaptive configurations. Results show LiDAR dominates near-field accuracy, the camera sustains longer-range coverage when LiDAR becomes unavailable, and the adaptive policy achieves a favorable accuracy-continuity trade-off by switching modalities based on information gain. By avoiding continuous multi-stream processing, the adaptive configuration provides a practical baseline for resilient and resource-aware maritime surveillance.

Adaptive Entropy-Driven Sensor Selection in a Camera-LiDAR Particle Filter for Single-Vessel Tracking

Abstract

Robust single-vessel tracking from fixed coastal platforms is hindered by modality-specific degradations: cameras suffer from illumination and visual clutter, while LiDAR performance drops with range and intermittent returns. We present a heterogeneous multi-sensor fusion particle-filter tracker that incorporates an information-gain (entropy-reduction) adaptive sensing policy to select the most informative configuration at each fusion time bin. The approach is validated in a real maritime deployment at the CMMI Smart Marina Testbed (Ayia Napa Marina, Cyprus), using a shore-mounted 3D LiDAR and an elevated fixed camera to track a rigid inflatable boat with onboard GNSS ground truth. We compare LiDAR-only, camera-only, all-sensors, and adaptive configurations. Results show LiDAR dominates near-field accuracy, the camera sustains longer-range coverage when LiDAR becomes unavailable, and the adaptive policy achieves a favorable accuracy-continuity trade-off by switching modalities based on information gain. By avoiding continuous multi-stream processing, the adaptive configuration provides a practical baseline for resilient and resource-aware maritime surveillance.
Paper Structure (32 sections, 28 equations, 5 figures, 1 table, 1 algorithm)

This paper contains 32 sections, 28 equations, 5 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. State Vector
  4. Boat Motion Model
  5. Observation Models
  6. Camera
  7. LiDAR
  8. Sensor Configuration
  9. Methodology
  10. Spatial Calibration to the World Frame
  11. Temporal Calibration (GNSS-Camera-LiDAR)
  12. Camera Detection Pipeline
  13. LiDAR Detection Pipeline
  14. Multi-Modal Particle Filter
  15. Initialization
  16. ...and 17 more sections

Figures (5)

  • Figure 1: Multi-sensor platform (Camera and LiDAR) deployed at a coastal location at Ayia Napa Marina.
  • Figure 2: (a) Camera frame with detected bounding box. (b) Dual-view LiDAR visualization: a spherical intensity map with a bounding box overlay (left), and a Bird’s Eye View plotting the target's specific range radius and azimuth bearing (right).
  • Figure 3: System overview showing parallel sensor processing streams converging to the particle filter for multi-modal fusion.
  • Figure 4: Map context showing the three evaluation zones (Zone 1: marina, Zone 2: near, Zone 3: far) and the GNSS reference trajectory; zone assignment is based on the GNSS position.
  • Figure 5: Zone-wise tracking performance across fusion regimes: RMSE (A) between the particle-filter position estimate and GNSS ground truth; Lost-bin percentage by zone (B). Markers denote mean, error bars indicate 95% confidence intervals.