Monocular Unmanned Boat Ranging System Based on YOLOv11-Pose Critical Point Detection and Camera Geometry

Yuzhen Wu, Yucheng Suo, Xinqiang Chen, Yongsheng Yang, Han Zhang, Zichuang Wang, Octavian PostolacheJournal of Marine Science and Engineering 2025 · DOI

TL;DR

This paper introduces a novel monocular vision system for unmanned boats that uses YOLOv11m-pose to detect critical key points on the boat. By combining these key points with camera geometry and calibration, the system achieves more accurate and robust distance estimation compared to traditional bounding box methods, especially in challenging marine environments.

What Problem Does This Paper Address?

Autonomous navigation for unmanned boats heavily relies on accurately sensing their surroundings, with distance detection being a crucial component. While monocular vision offers advantages like low cost and simple deployment, achieving precise and reliable distance measurements from a single camera, especially in dynamic and complex marine conditions, remains a challenge. Traditional methods often struggle with factors like varying lighting, water reflections, and the inherent limitations of bounding box detection, leading to significant errors. This paper aims to improve the accuracy and robustness of monocular ranging for unmanned boats by leveraging a key point detection approach.

Methods

The core of this system is a two-stage process:

  1. Key Point Detection: The researchers employ YOLOv11m-pose, a state-of-the-art object detection model enhanced for key point estimation. This model is specifically adapted to identify and locate seven critical key points on the unmanned boat's hull. The YOLOv11m-pose architecture incorporates novel modules like C3K2 for multi-channel feature fusion and C2PSA for dynamic attention, which are designed to improve feature extraction speed and accuracy, particularly in complex scenes.
  2. Physical Distance Mapping: Once the key points are detected, their pixel coordinates are used in conjunction with camera calibration parameters (obtained using Zhang's calibration method) to estimate the real-world distance. This involves:
  3. Camera Calibration: Determining intrinsic parameters like focal length and principal point coordinates.
  4. Geometric Projection: Utilizing principles of triangular similarity and camera geometry. The distance is calculated by relating the pixel coordinates of the key points to the camera's focal length, pitch angle, and the known height of the camera from the water surface. A key point closest to the water surface is specifically chosen for more accurate mapping.

This approach contrasts with traditional methods that rely on the bounding box of the detected object, which can be less precise due to factors like partial occlusion or inaccurate box fitting.

Datasets

The experiments utilized real navigation video data of an unmanned boat model.

  • Training and Validation: A total of 996 images were extracted from the video, with 712 used for training and 284 for validation.
  • Real-World Ranging Samples: A dataset of 76 typical representative real unmanned boat ranging samples was collected across four scenarios. The real distance was measured using a laser rangefinder.

The paper does not explicitly mention the creation of a new benchmark dataset but rather uses collected real-world data to evaluate their proposed method against existing approaches.

Key Insights and Ideas

  • Key Points Over Bounding Boxes: The most significant insight is that using specific, well-defined key points on the unmanned boat provides a more robust and accurate basis for distance estimation than relying solely on the object's bounding box. This is because key points are less susceptible to variations in lighting, reflections, and partial occlusions that can distort bounding boxes.
  • YOLOv11m-pose's Efficacy: The YOLOv11m-pose model demonstrates strong performance in detecting these critical key points, outperforming other YOLO-pose variants in key point detection metrics. This highlights its suitability for complex marine environments.
  • Environmental Robustness: The key point-based method shows significantly better performance across various challenging scenarios, including strong light, low light, and foggy conditions, compared to the detection frame-based method. This underscores its practical applicability in real-world maritime navigation.
  • Camera Geometry is Crucial: The paper emphasizes the importance of accurate camera calibration and geometric principles in translating 2D pixel information into 3D real-world distances.

Key Results

  • Key Point Detection Accuracy: YOLOv11m-pose achieved high scores in Precision-Pose (0.940), Recall-Pose (0.845), mAP50-Pose (0.898), and mAP50-95-Pose (0.757), demonstrating its effectiveness for unmanned boat key point detection.
  • Distance Mapping Accuracy: The key point-based distance mapping method consistently outperformed the detection frame-based method across various scenarios. For instance, in Scene I, the MSE was reduced from 0.395 to 0.284, and MAE from 0.508 to 0.416. In challenging foggy conditions (Scene V and VI), the key point method also showed substantial improvements in MSE, RMSE, and MAE.
  • Error Reduction: The proposed method significantly reduces distance measurement errors, particularly in adverse conditions like strong light with water reflections, where it achieved an error rate of 1.4% compared to 7.2% for the detection frame method.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that the accuracy of the distance mapping is sensitive to the accuracy of the camera focal length parameter.
  • While robust, the method's performance can still be affected by extreme environmental conditions that severely degrade feature extraction.
  • Future Work:
    • The authors suggest directing future research towards more complex autonomous sensing for unmanned ships.
  • Exploring more advanced sensor fusion techniques or multi-camera systems could further enhance robustness and accuracy.
  • Investigating methods to dynamically adapt to changing environmental conditions or camera parameters in real-time could be beneficial.

Summary updated

Abstract

Unmanned boat distance detection is an important foundation for autonomous navigation tasks of unmanned boats. Monocular vision ranging has the advantages of low hardware equipment requirements, simple deployment, and high efficiency of distance detection. Unmanned boats can sense the real-time navigational situation of waters through monocular vision ranging, providing data support for their autonomous navigation. This paper establishes a framework for unmanned boat distance detection. The framework extracts and recognizes the features of an unmanned boat through Yolov11m-pose and selects the key points of the ship for physical distance mapping. Using the camera calibration to obtain the pixel focal length, the main point coordinates and other parameters are obtained. The number of pixel points in the image key point to the image center pixel and the actual distance of the camera from the horizontal plane are combined with the focal length of the camera for triangular similarity conversion. These data are fused with the camera pitch angle and other parameters to obtain the final distance. At the same time, experimental verification of the key point detection model demonstrates that it fully meets the requirements for unmanned boat ranging tasks, as assessed by Precision, Recall, mAP50, mAP50-95 and other indicators. These indicators show that Yolov11m-pose has a better accuracy in the key point detection task with an unmanned boat. The verification experiments also illustrate the accuracy of the key point-based physical distance mapping compared with the traditional detection frame-based physical distance mapping, which was assessed by the mean squared error (MSE), the root mean square error (RMSE), and the mean absolute error (MAE). The metrics show that key point-based unmanned boat distance mapping has greater accuracy in a variety of environmental situations, which verifies the effectiveness of this approach.
BibTeX
@article{wu2025monocular,
  author = {Yuzhen Wu and Yucheng Suo and Xinqiang Chen and Yongsheng Yang and Han Zhang and Zichuang Wang and Octavian Postolache},
  title = {{Monocular Unmanned Boat Ranging System Based on YOLOv11-Pose Critical Point Detection and Camera Geometry}},
  journal = {Journal of Marine Science and Engineering},
  year = {2025},
  doi = {10.3390/jmse13061172},
}

Loading related content...

Figure