The rapid growth of underwater exploration technologies has increased the need for precise detection of underwater obstacles, especially in shallow waters. Airborne lidar-equipped drones have emerged as a promising solution due to their ability to quickly and precisely assess underwater conditions. However, traditional three-dimension (3D) point clouds modeling methods, while effective in depicting the spatial relationships between the water surface, seabed, and obstacles, face challenges in terms of computational efficiency and rely heavily on precise measurements of water parameters. To address these limitations, we propose a novel image processing-based approach for airborne lidar data, which enhances the accuracy of underwater obstacle detection, especially in shallow water regions, where computational speed and precision are critical for safety and navigation.

We employ a comprehensive image processing framework, begining with the preprocessing of lidar waveform data, to address common challenges in underwater lidar applications. Specifically, the waveform data is processed to mitigate the effects of backscatter from the seawater and strong reflection areas caused by the angle of incidence of the laser on the water surface. This is accomplished by using linear interpolation to reduce the broadening effect of the water surface and applying a local maximum gradient method to remove bright streaks generated by specular reflections. After preprocessing, the waveforms are converted into composite images where obstacle detection is performed using a multi-scale difference of Gaussian (DoG) enhancement technique combined with contrast-based discrimination (DoG-ECD). This method enables efficient and accurate obstacle identification, even in areas affected by low contrast or background interference. Additionally, the detection process integrates adaptive thresholding based on the Sauvola algorithm, optimizing the thresholding process based on local pixel statistics. A scoring mechanism based on the largest connected domain area and the total number of connected domains further enhances the identification process. This method also accounts for varying obstacle sizes by applying multi-scale processing, selecting the best scale that emphasizes the most significant obstacle feature, ensuring both high detection reliability and computational efficiency. This high robust and scalable approach makes the method highly effective for diverse underwater environments.

Experimental validation is conducted using real-world data collected from a UAV (Unmanned Aerial Vehicle)-mounted lidar system deployed in Qiandao Lake, covering a range of underwater scenes to test the proposed method under diverse conditions. Results demonstrate that the method significantly improves the accuracy of underwater obstacle detection, achieving an overall accuracy of 94.55% and an F1-score of 94.97%. The DoG enhancement technique is particularly effective in amplifying the contrast between obstacles and surrounding water, even in the presence of noise and surface reflections, ensuring that obstacles are clearly distinguishable. The contrast-based discrimination method plays a crucial role in identifying true obstacles by differentiating them from background noise. It minimizes false positives (incorrectly identifying background noise as obstacles) while maximizing true positives (accurately detecting actual obstacles), which is crucial in real-world applications where false positives can lead to unnecessary delays and interventions. In comparison to traditional detection methods, such as the Canny edge detection algorithm, the proposed method outperforms it, particularly in detecting obstacles in full images (Fig. 7). The proposed method achieved a 31-Percentage-point improvement in accuracy, reaching 93.4% (Table 2). Furthermore, the proposed method demonstrates superior performance in processing large datasets, making it highly suitable for real-time applications in trajectory planning and obstacle detection, where fast and accurate identification is essential for operational success.

We introduce a novel image processing approach that significantly enhances the performance of airborne lidar systems for underwater obstacle detection. By addressing challenges related to scattering effects, water surface reflections, and low contrast between obstacles and their background, the proposed method offers a more efficient and accurate solution for detecting underwater obstacles in shallow waters. The method’s high accuracy and robust performance make it a valuable tool for a variety of applications, including underwater exploration, navigation planning, and marine resource development. Compared to traditional 3D point cloud modeling methods, which are computationally expensive and time-consuming, the proposed method offers a more scalable and efficient alternative that can be applied in real-time scenarios. In addition, the method’s ability to handle large volumes of data makes it a promising tool for future marine technology applications.