Benchmarking Online Object Trackers for Underwater Robot Position Locking Applications

TL;DR

This work addresses underwater vision-based ROV position locking by benchmarking seven one-shot trackers (Boosting, MIL, MedianFlow, MOSSE, TLD, KCF, CSRT) within a closed perception-control loop using a PI controller. It integrates tracker outputs with real-time control on a BlueROV2 in an indoor pool and provides extensive real-world results across disturbance scenarios, identifying CSRT as the most robust choice for underwater applications. The authors also release an open-source underwater ROV dataset to facilitate future research and benchmarking. Overall, the study demonstrates that CSRT consistently delivers the best trade-off between position accuracy and yaw stability, making it a strong candidate for practical underwater vision-based control systems.

Abstract

Autonomously controlling the position of Remotely Operated underwater Vehicles (ROVs) is of crucial importance for a wide range of underwater engineering applications, such as in the inspection and maintenance of underwater industrial structures. Consequently, studying vision-based underwater robot navigation and control has recently gained increasing attention to counter the numerous challenges faced in underwater conditions, such as lighting variability, turbidity, camera image distortions (due to bubbles), and ROV positional disturbances (due to underwater currents). In this paper, we propose (to the best of our knowledge) a first rigorous unified benchmarking of more than seven Machine Learning (ML)-based one-shot object tracking algorithms for vision-based position locking of ROV platforms. We propose a position-locking system that processes images of an object of interest in front of which the ROV must be kept stable. Then, our proposed system uses the output result of different object tracking algorithms to automatically correct the position of the ROV against external disturbances. We conducted numerous real-world experiments using a BlueROV2 platform within an indoor pool and provided clear demonstrations of the strengths and weaknesses of each tracking approach. Finally, to help alleviate the scarcity of underwater ROV data, we release our acquired data base as open-source with the hope of benefiting future research.

Figures (16)

• Figure 1: Benchmarking online tracking algorithms for underwater position locking. a) the Blue Robotics BlueROV2 robot used in our experiments; b) An underwater structure is used as an object of interest to be tracked by the ROV; c) the ROV is equipped with a front-facing camera which is used by the object tracker under test to detect and track objects as part of the position locking system; d) a pole is used to generate arbitrary disturbances on the ROV position in order to study the robustness of the object trackers during our experiments.
• Figure 2: Block diagram of the ROV control setup. The ROV platform is connected via a tether to a USB interface. The laptop running the position control script receives the ROV camera feed and IMU readings via the tether USB interface and sends control commands to the ROV using this same interface. Optionally, an Xbox controller can be used for the manual control of the ROV for initiating its position in the pool (not used during our automated control experiments).
• Figure 3: Types of objects to be tracked for position locking. a) structure object b) ladder object.
• Figure 4: Position locking error-Structure-No disturbances.
• Figure 5: Yaw angle error-Structure-No disturbances.