Towards High-Resolution Seafloor 3D Mapping from Underwater Flash LiDAR and ICP Registration

HAL Id: hal-05094692 https://hal.science/hal-05094692v1 Submitted on 6 Jun 2025 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. Distributed under a Creative Commons Attribution 4.0 International License Towards High-Resolution Seafloor 3D Mapping from Underwater Flash LiDAR and ICP Registration Maxime Ferrera, Christophe Pache, Jannis Holzer, Eleonoor Bosch, Aurélien Arnaubec, Bernard Puigserver, Rahima Djahel, Javier Escartín, Jan Opderbecke To cite this version: Maxime Ferrera, Christophe Pache, Jannis Holzer, Eleonoor Bosch, Aurélien Arnaubec, et al.. Towards High-Resolution Seafloor 3D Mapping from Underwater Flash LiDAR and ICP Registration. OCEANS 2025 - Brest, Jun 2025, Brest, France. �hal-05094692� Towards High-Resolution Seafloor 3D Mapping from Underwater Flash LiDAR and ICP Registration Maxime Ferrera1, Christophe Pache2, Jannis Holzer2, Eleonoor Bosch2, Aur´ elien Arnaubec1, Bernard Puigserver1, Rahima Djahel3, Javier Escartin3, Jan Opderbecke1 Abstract—Seafloor mapping is crucial for ocean exploration. Current techniques mostly involve either long-range acoustic sensors or short-range passive optical systems. This paper presents an innovative underwater flash lidar system that bridges the gap by enabling high-resolution 3D mapping at intermediate ranges. To ensure accurate registration of lidar measurements, we propose an automatic pipeline based on the Iterative Closest Point (ICP) algorithm. Preliminary field experiments demonstrate the system’s potential for applications in geosciences, marine biology, archaeology and infrastructure monitoring, highlighting underwater lidar as a promising tool for detailed and efficient seafloor mapping. Index Terms—Underwater, Robotics, Lidar, ICP, 3D Mapping. I. INTRODUCTION Seafloor mapping is a critical component of ocean exploration. While ship-mounted multi-beam acoustic sensors can be used for this purpose, the resolution of the resulting maps is often limited due to the significant distance between the ship and the seafloor. Autonomous Underwater Vehicles (AUVs) and Remotely Operated Vehicles (ROVs) are therefore preferred for high-resolution mapping, as they can operate in close proximity to the seabed. Seafloor mapping with ROVs and AUVs primarily relies on two methods: passive optical mapping and acoustic systems [1]. Optical mapping, conducted at close range (<10m) due to illumination constraints, achieves high resolutions of approximately 1 cm [2]–[4]. In contrast, acoustic systems operate at longer ranges (>50 m) but provide coarser resolutions (>20 −50 cm). Thus, these methods are complementary in terms of resolution and coverage. Underwater lidar potentially bridges the gap between them, offering intermediate survey ranges and resolutions while enabling efficient large-area mapping when integrated into deep-sea vehicles. The potential applications of such a system underwater are similar to landor aerial-based systems, but untapped to date. This includes This work benefits from a French government grant managed by the National Research Agency (ANR) under the Programme d’Investissements d’Avenir integrated into France 2030, bearing the reference ANR-21-ESRE0042. This research is also co-funded by the European Union’s Horizon 2020 research and innovation program under Grant Agreement No 101094818 (https://triquetra-project.eu). 1Ifremer, SM, RI French Oceanographic Fleet, F-83500 La Seyne-sur-Mer, France. Contact email: {[email protected]} 2Centre Suisse d’Electronique et de Microtechnique SA (CSEM), CH-2002 Neuchˆ atel, Switzerland. Contact email: {[email protected]} 3Laboratoire de G´ eologie, CNRS UMR8538, ENS, PSL University, 75231 Paris, France. Contact email: {[email protected], [email protected]} applications in geosciences (e.g. mapping of outcrops and the study of processes such as sedimentation and its dynamics, characterization of ground deformation) [5], biology (mapping of habitats in high-resolution, monitoring of biomass evolution) [6], infrastructure survey (effects of their installation on the seabed, health monitoring), archaeology and preservation of cultural heritage [7], [8]. In this paper, we present an innovative underwater lidar system that aims at generating 3D maps of scanned environments. The generation of such 3D maps is usually done from dead-reckoning navigation using an Inertial Navigation System (INS) coupled with a Doppler Velocity Log (DVL) when performing acoustic surveys at long range [9]. However, INS systems are prone to cumulative errors and drift which can lead to misaligned successive point clouds, especially at closer range [10], [11]. The navigation thus requires to be refined for optical systems. For instance, when doing photogrammetry from images or videos, structure-from-motion techniques must be applied to estimate a more accurate navigation from optical data [12]–[17]. In the same way, lidar measurements will require a refined navigation to be accurately registered [18], [19]. To address this issue, we explore the use of the Iterative Closest Point (ICP) algorithm [20] to refine navigation estimates and improve the alignment of lidar-generated data. The contributions of this paper are the following: •the description of a flash lidar system suited to underwater environments •an algorithmic pipeline to accurately register the lidar measurements in a common 3D point cloud The paper is organized as follows. We first describe the developed lidar system in section II. Then, we describe the ICP-based algorithmic pipeline that automatically register lidar measurements in section III. Finally, we present results obtained during a field experiment with a prototype of the lidar in a lake in section IV. II. UNDERWATER FLASH LIDAR TECHNOLOGY Over the years, several techniques have been developed for three-dimensional optical imaging for subsea applications to benefit from an increased resolution [21]. Nevertheless, imaging in attenuating underwater environments remains a major challenge for the community [22]. Even with artificial illumination, in turbid water conditions, cameras are saturated by scattered light coming from the particles in suspension. In clear water, photogrammetry still faces calibration challenges, limiting its applicability to close-range inspections (typically below 5 m) [23]. To reach longer distances, sensors based on active illumination combined with time-gating are thus required. Lidar is a cutting-edge technology for mapping distances using light time-of-flight (TOF) measurements and is used for 3D imaging of static scenes to create high-resolution maps or for tracking moving targets. Apart from interferometric methods, there exist two main techniques [24]: •Indirect TOF uses continuous sinusoidal modulation to measure phase shifts between reference and reflected signals. •Direct TOF measures the time a laser pulse takes to travel to and from a target, directly relating to distance. Indirect TOF suffers from limitations: it cannot resolve multiple echoes and is sensitive to background light, making it less suitable for outdoor use. Recent advancements thus favour direct TOF detectors. Lidar systems can be either scanning or flash. Scanning lidar captures one or several pixels at a time with a collimated, high-power laser beam, while flash lidar illuminates a large area to gather data from many pixels simultaneously [21]. For the same power consumption, flash lidar is less effective over a long range because it distributes light over a large FOV, reducing light intensity and maximum distance. However, flash lidar benefits from key advantages: •High reliability due to the absence of a scanning mechanism, reducing mechanical fatigue and allowing for miniaturization. •3D images are captured in a single snapshot, minimizing motion blur. •Fixed angular resolution across the entire FOV, with resolution limited by the sensor rather than scanning stability. These features make flash lidar ideal for niche markets with limited solutions. Green wavelength lidar products designed for underwater imaging exist on the market but these systems either lack miniaturization or are based on a scanning system, making them suboptimal for moving platforms. In comparison, thanks to the absence of scanning mechanisms and snapshot acquisition, flash lidar systems can be integrated onto autonomous platforms [25]. Initially developed for space applications [26], the described flash lidar technology is also suitable for underwater imaging [27] due to its use of green lasers, which align with the TOF detector’s highest sensitivity [28], developed by the Fondazione Bruno Kessler (Italy). The current lidar prototype shown in Figure 1 weighs 6.8 kg and holds within a volume of 20 x 17 x 19 cm – without its pressure-tolerant underwater housing – with a power consumption of 60 W in operation. The system’s core consists in a state-of-the-art embedded processing unit (Xlinx Zynq Ultrascale+) that allows controlling the data acquisition, the actuators positioning and communication. Based on SPADs, each pixel precisely time stamps detection events thanks to an individual counter and Fig. 1. CSEM Airswim flash lidar prototype. a time-to-digital converter. The processing unit also acts as a master for the fibre amplifier laser that emits nanosecond pulses with high energy (10 µJ). Due to the inherent noise of SPADs, in order to acquire a single 3D image with sufficient quality, several measurements are acquired. Each measurement is synchronized onto a single laser pulse. By summing measurements, a histogram of photon counts is built with respect to the distance for each pixel. A depth map (i.e. 3D image), as shown in Figure 2, is finally computed by identifying the peak position in the histogram. Fig. 2. Left: Picture of a test scene with lidar illumination. Right: Lidar acquired depth map. III. UNDERWATER LIDAR 3D MAPPING In this paper, we aim at producing accurate 3D maps of underwater environments from the lidar sensor described in the previous section. This lidar is going to be carried by ROV or AUV systems while they navigate close to the seafloor. The lidar acquisitions are based on precise TOF measurements from the emission of the laser beam until the reception of its reflection on scanned structures. These TOF measurements are then to be converted to range measurements in order to produce 3D point clouds. Hence, at each acquisition, the lidar produces a 3D point cloud and each of the produced point clouds has to be registered with respect to the previous one in order to create an accurate 3D map. We now detail our pipeline from the pre-processing of lidar TOF measurements into range values to the accurate registration of multiple lidar acquisitions. A. Preprocessing of the lidar measurements The underwater lidar system consists of a laser emitter and a receiver, each placed behind a flat port. The receiver is made of an optical lens system that focuses the received laser light reflections on a dedicated matrix of pixel-like sensors. Similarly to conventional cameras, we can geometrically describe the lidar receiver using intrinsic parameters with f∈Rdenoting the focal length and (cx, cy)∈R2the principal point [29]. For simplicity, we make the following assumptions on our systems: •We neglect the very short traveling of light in air and in the flat port material •We consider an ideal optical lens with no distortion Then, for each lidar acquisition, we have one Time-of-Flight (TOF) measurement ∆tper pixel. We can thus convert each of these measurements into a range value: r=c nwater ·∆t 2.(1) where cis the speed of light in air, and nwater is the refractive index of water. To finally produce a point cloud from the range values, we compute the incoming ray rair ∈R3for each pixel as follows: rair =u−cx f,v−cy f,1.(2) with (u, v)∈R2the pixel coordinates from which the ray originates. We then apply Snell’s law to compute the refracted in-water ray: rwater =s1−nair nwater 21−(n·rair)2n +nair nwater (rair −(n·rair)n). (3) Taking the unit-norm version of the in-water ray, we finally multiply it by its associated range value to obtain a 3D point: X=r·rwater ∥rwater∥.(4) Note that each 3D point X∈R3is computed in the lidar frame. We differentiate these 3D points based on their acquisition time, now denoting points measured during an acquisition at t=ias Xi j(i.e., Xi jis a 3D point computed in the lidar frame at time i). To produce consistent 3D maps from such 3D points, we next have to project each of them into a world frame. B. Registration of lidar 3D point clouds Each lidar acquisition produces a set of 3D points {Xi j}. To register these points in the world frame, we need to know the 3D rigid Euclidean transformation that links the lidar at t=i to this world frame. That is, we need to know the pose of the lidar at t=i: Xw j=Rwi ·Xi j+twi (5) where Rwi ∈SO(3) and twi ∈R3are respectively the rotation matrix and the translation vector representing the pose of the lidar at i. In underwater environments, vehicles’ poses are typically estimated using Inertial Navigation Systems (INS) coupled with a Doppler Velocity Log (DVL). However, such systems are prone to cumulative errors and drift, which will eventually lead to misaligned point clouds. To address this issue, we explore the use of Iterative Closest Point (ICP) techniques [20], [30] to automatically register successively acquired point clouds. We employ the Point-to-Plane variant of the classical ICP [31] : arg min Rwi,twi X (p,q)∈L ∥(p−(Rwi ·q+twi)) ·np∥φ(6) Where npis the normal of point p,Lis the correspondence set between the target point cloud P, and source point cloud Q, and φdenotes the Cauchy loss. In order to be robust to potential outliers, we use a Cauchy Loss φwhen solving the above equation, thus applying an Iterative Re-weighted Least Square (IRLS) approach [32]. Like classical Non-Linear Least Square methods based on the Gauss-Newton or Levenberg-Marquardt algorithms, IRLS is a local optimization technique. To avoid starting from too far from the solution and possibly increase the convergence rate, we also initialize (Rwi,twi)using a constant velocity motion model and the last pose estimate. We note that, the lidar described in this paper still being at a prototypal stage, we have not yet tested it with an INS-DVL equipped underwater vehicle. In such a case, it would be interesting to replace the constant velocity motion model with actual INS navigation estimates. IV. FIELD EXPERIMENT RESULTS The ICP-based registration pipeline has been developed in Python and relies on the ICP implementation of the Open3D library [33]. Each lidar acquisition is first statistically filtered by removing points that are farther away than the average distance plus one standard deviation between their 16 nearest neighbors. The robust Cauchy loss threshold is set to 0.025 m in the Point-to-Plane ICP. The method proposed in this paper has been successfully applied on a trial dataset acquired with an Unmanned Surface Vehicle (USV) operating on a lake, as shown in Figure 3. During the trial, the lidar sensor was submerged to map the seabed at a depth of 2 meters, while GPS and IMU data Fig. 3. BathyDrone Sumo operating in Lake Neuchˆ atel and the lidar mounted in its waterproof case (top right) Fig. 4. Lidar based 3D map from USV on-board GPS and IMU navigation data. Fig. 5. Lidar based 3D map registered with the Iterative Closest Point algorithm. were simultaneously recorded. The dataset comprises 100 lidar measurements paired with navigation data. Figure 4 presents the results of point cloud projection using navigation estimates derived directly from the GPS and IMU data, revealing noticeable misalignment between successive point clouds. In contrast, Figure 5 shows the results obtained with our ICP-based pipeline, demonstrating significantly improved alignment. These results highlight the potential of the ICP approach in enhancing the accuracy and consistency of lidar-based 3D maps, paving the way for reliable seafloor mapping. V. CONCLUSION In this paper, we have introduced an innovative underwater flash lidar system designed to enhance seafloor mapping by bridging the gap between short-range passive optical and longrange acoustic methods. By integrating a lidar into underwater vehicles, we enable high-resolution 3D mapping at intermediate ranges, offering a valuable tool for various scientific and industrial applications. To address the challenge of accurate data registration, we proposed an algorithmic pipeline leveraging the Iterative Closest Point (ICP) algorithm to refine navigation estimates and align lidar-generated point clouds. Our preliminary field experiments demonstrate the potential of this approach, showing improved accuracy in underwater mapping. This system opens new possibilities for applications in geosciences, marine biology, infrastructure monitoring and underwater archaeology, providing a more efficient and precise means of surveying underwater environments. Future work will focus on miniaturizing the current lidar prototype and further optimizing the registration pipeline through Simultaneous Localization And Mapping (SLAM). The lidar presented in this work targets deep-sea science and is part of an extensive set of multidisciplinary scientific equipments developed in the DeepSea’Nnovation [34] project for the 6000m depth ranged underwater vehicles operated within the French Oceanographic Fleet. REFERENCES [1] J. J. Leonard and A. Bahr, “Autonomous underwater vehicle navigation,” Springer handbook of ocean engineering, pp. 341–358, 2016. [2] N. Gracias and J. Santos-Victor, “Underwater video mosaics as visual navigation maps,” Computer Vision and Image Understanding, vol. 79, no. 1, pp. 66–91, 2000. [3] M. Ferrera, V. Creuze, J. Moras, and P. Trouv´ e-Peloux, “Aqualoc: An underwater dataset for visual–inertial–pressure localization,” The International Journal of Robotics Research, vol. 38, no. 14, pp. 1549– 1559, 2019. [4] A. Arnaubec, M. Ferrera, J. Escartin, M. Matabos, N. Gracias, and J. Opderbecke, “Underwater 3d reconstruction from video or still imagery: Matisse & 3dmetrics processing and exploitation software,” Journal of Marine Science and Engineering, 2023. [5] N. G. Pricope and M. S. Bashit, “Emerging trends in topobathymetric lidar technology and mapping,” International Journal of Remote Sensing, vol. 44, no. 24, pp. 7706–7731, 2023. [6] D. W. Caress, E. J. Martin, M. Risi, G. Troni, A. Hamilton, C. Kecy, J. B. Paduan, H. J. Thomas, S. M. Rock, M. Wolfson-Schwehr, et al., “The mbari low-altitude survey system for 1-cm-scale seafloor surveys in the deep ocean,” IEEE Journal of Oceanic Engineering, 2025. [7] N. Gracias, P. Ridao, R. Garcia, J. Escart´ ın, M. L’Hour, F. Cibecchini, R. Campos, M. Carreras, D. Ribas, N. Palomeras, et al., “Mapping the moon: Using a lightweight auv to survey the site of the 17th century ship ‘la lune’,” in 2013 MTS/IEEE OCEANS-Bergen, pp. 1–8, IEEE, 2013. [8] O. Khatib, X. Yeh, G. Brantner, B. Soe, B. Kim, S. Ganguly, H. Stuart, S. Wang, M. Cutkosky, A. Edsinger, et al., “Ocean one: A robotic avatar for oceanic discovery,” IEEE Robotics & Automation Magazine, vol. 23, no. 4, pp. 20–29, 2016. [9] J. C. Kinsey, R. M. Eustice, and L. L. Whitcomb, “A survey of underwater vehicle navigation: Recent advances and new challenges,” in IFAC conference of manoeuvering and control of marine craft, vol. 88, pp. 1–12, Lisbon, 2006. [10] G. Inglis, C. Smart, I. Vaughn, and C. Roman, “A pipeline for structured light bathymetric mapping,” in 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 4425–4432, IEEE, 2012. [11] M. Ferrera, A. Arnaubec, K. Isteniˇ c, N. Gracias, and T. Bajjouk, “Hyperspectral 3d mapping of underwater environments,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 3703– 3712, 2021. [12] O. Pizarro and H. Singh, “Toward large-area mosaicing for underwater scientific applications,” IEEE journal of oceanic engineering, vol. 28, no. 4, pp. 651–672, 2003. [13] O. Pizarro, R. M. Eustice, and H. Singh, “Large area 3-d reconstructions from underwater optical surveys,” IEEE Journal of Oceanic Engineering, vol. 34, no. 2, pp. 150–169, 2009. [14] Ferrera, Maxime and Moras, Julien and Trouv´ e-Peloux, Pauline and Creuze, Vincent, “Real-time monocular visual odometry for turbid and dynamic underwater environments,” Sensors, vol. 19, no. 3, p. 687, 2019. [15] C. Boittiaux, C. Dune, M. Ferrera, A. Arnaubec, R. Marxer, M. Matabos, L. Van Audenhaege, and V. Hugel, “Eiffel tower: A deep-sea underwater dataset for long-term visual localization,” The International Journal of Robotics Research, vol. 42, no. 9, pp. 689–699, 2023. [16] M. Ferrera, A. Arnaubec, C. Boittiaux, I. Larroche, and J. Opderbecke, “Vision-based 3d reconstruction for deep-sea environments: Practical use for surveys and inspection,” in OCEANS 2023-Limerick, pp. 1–7, IEEE, 2023. [17] M. She, Y. Song, D. Nakath, and K. K¨ oser, “Semihierarchical reconstruction and weak-area revisiting for robotic visual seafloor mapping,” Journal of Field Robotics, vol. 41, no. 8, pp. 2749–2770, 2024. [18] M. Massot-Campos, G. Oliver, A. Bodenmann, and B. Thornton, “Submap bathymetric slam using structured light in underwater environments,” in 2016 IEEE/OES Autonomous Underwater Vehicles (AUV), pp. 181–188, IEEE, 2016. [19] A. Palomer, P. Ridao, and D. Ribas, “Inspection of an underwater structure using point-cloud slam with an auv and a laser scanner,” Journal of field robotics, vol. 36, no. 8, pp. 1333–1344, 2019. [20] P. J. Besl and N. D. McKay, “Method for registration of 3-d shapes,” in Sensor fusion IV: control paradigms and data structures, vol. 1611, pp. 586–606, Spie, 1992. [21] M. Castill´ on, A. Palomer, J. Forest, and P. Ridao, “State of the art of underwater active optical 3d scanners,” Sensors, vol. 19, no. 23, p. 5161, 2019. [22] A. Filisetti, A. Marouchos, A. Martini, T. Martin, and S. Collings, “Developments and applications of underwater lidar systems in support of marine science,” in OCEANS 2018 MTS/IEEE Charleston, pp. 1–10, IEEE, 2018. [23] M. Massot-Campos and G. Oliver-Codina, “Optical sensors and methods for underwater 3d reconstruction,” Sensors, vol. 15, no. 12, pp. 31525– 31557, 2015. [24] M.-C. Amann, T. M. Bosch, M. Lescure, R. A. Myllylae, and M. Rioux, “Laser ranging: a critical review of unusual techniques for distance measurement,” Optical engineering, vol. 40, pp. 10–19, 2001. [25] P. Risholm, J. Thorstensen, J. T. Thielemann, K. Kaspersen, J. Tschudi, C. Yates, C. Softley, I. Abrosimov, J. Alexander, and K. H. Haugholt, “Real-time super-resolved 3d in turbid water using a fast range-gated cmos camera,” Applied optics, vol. 57, no. 14, pp. 3927–3937, 2018. [26] A. Pollini, C. Pache, and J. Haesler, “Csem space lidars for imaging and rangefinding,” in IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium, pp. 1849–1852, IEEE, 2018. [27] C. Pache, C. Meier, S. Droz, D. Nguyen, J.-C. Roulet, A. Pollini, T. Hauge, and F. Droz, “Miniature flash lidar for bathymetry: An experimental proof-of-concept,” in 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, pp. 8531–8533, IEEE, 2021. [28] M. Perenzoni, D. Perenzoni, and D. Stoppa, “A 64x64-pixels digital silicon photomultiplier direct tof sensor with 100-mphotons/s/pixel background rejection and imaging/altimeter mode with 0.14% precision up to 6 km for spacecraft navigation and landing,” IEEE Journal of Solid-State Circuits, vol. 52, no. 1, pp. 151–160, 2016. [29] Hartley, Richard and Zisserman, Andrew, Multiple view geometry in computer vision. 2003. [30] S. Rusinkiewicz and M. Levoy, “Efficient variants of the icp algorithm,” in Proceedings third international conference on 3-D digital imaging and modeling, pp. 145–152, IEEE, 2001. [31] Y. Chen and G. Medioni, “Object modelling by registration of multiple range images,” Image and vision computing, vol. 10, no. 3, pp. 145–155, 1992. [32] P. Babin, P. Giguere, and F. Pomerleau, “Analysis of robust functions for registration algorithms,” in 2019 International conference on robotics and automation (ICRA), pp. 1451–1457, IEEE, 2019. [33] Q.-Y. Zhou, J. Park, and V. Koltun, “Open3d: A modern library for 3d data processing,” arXiv preprint arXiv:1801.09847, 2018. [34] A.-G. Vincent, V. Chavagnac, P. Jaussaud, X. Saint-Laurent, E. Raugel, and J. Opderbecke, “The deepsea’nnovation project: Innovative sensing and sampling for breakthrough deep-sea marine sciences,” in IEEE OCEANS 2025, June 2025. In press.