Pith. sign in

REVIEW 3 major objections 5 minor 49 references

NavEYE fuses AIS, radar, and shipboard camera so intelligent surface vehicles get one track with identity, motion, and appearance.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

MCGA, distance-aware weighted fusion, time-decay stitching, and bearing-distance vision association improve multi-sensor ship tracking on a real shipborne MAPFusion dataset and a deployed NavEYE system.

T0 review reviewed 2026-07-11 challenge →

load-bearing objection Solid shipboard engineering paper: real multi-sensor voyage data, clear DAWF/TDSF switching, and a working NavEYE stack; fusion-error numbers rest mainly on one GoPro-instrumented ship, so treat the accuracy claims as scoped rather than general. the 3 major comments →

arxiv 2607.03915 v2 pith:3H3HBLMP submitted 2026-07-04 cs.CV

NavEYE: Vision-Centered Multi-Sensor Fusion-Based Situational Awareness System for Intelligent Surface Vehicles

Ryan Wen Liu , Junxiong Liang , Haoyu Wang , Mengwei Bao This is my paper
classification cs.CV
keywords situational awarenessmulti-sensor fusionvision-AIS-radar fusiondata associationintelligent surface vehiclesAIS-radar fusionship detection
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Intelligent surface vehicles need a continuous picture of nearby ships that carries identity, position and speed, and visual appearance at once. No single sensor delivers all three: AIS is sparse and can drop out, radar is noisy and distance-dependent, and a camera alone has no absolute motion state. This paper builds NavEYE, a vision-centered pipeline that first associates AIS with radar under multi-feature gating, then fuses them with distance-aware weights or time-decay stitching when a source disappears, and finally links the fused geospatial track to camera detections with joint normalized bearing and distance costs. On a real multi-scenario shipborne dataset, MAPFusion, the method reports higher association accuracy and smoother fused tracks than common baselines, and the shipboard system overlays MMSI, kinematics, and collision cues on live video and chart layers. A sympathetic reader cares because the claim is not a single algorithm score but an end-to-end path from raw AIS, radar, and video to navigational assistance that can reduce ambiguity and hard switches in dense traffic.

Core claim

On real synchronized AIS–radar–camera data, multi-constraint AIS–radar association (MCGA), distance-aware weighted fusion plus time-decay stitching (DAWF/TDSF), and joint bearing–distance vision association (VARF) produce more accurate, continuous fused ship tracks than standard association and fixed-weight or hard-switch fusion baselines, and support a working shipboard situational awareness system.

What carries the argument

The MCGA–DAWF/TDSF–VARF cascade: Mahalanobis multi-feature gated AIS–radar matching; adaptive fusion that reweights by distance-dependent radar error or stitches with exponential time decay when a source is lost; and Hungarian assignment on normalized bearing and image-vertical distance costs to attach visual detections to geospatial tracks.

Load-bearing premise

The method treats the bottom of a ship’s camera box as a reliable stand-in for how far away the ship is, which only holds if camera pitch and water-surface geometry stay stable enough for that vertical position to track true range.

What would settle it

On dense narrow-channel or port sequences with measured camera pitch change, wave occlusion, or strong scale variation, measure whether VARF association accuracy collapses when the vertical-box distance cue is removed or corrupted while bearing is held fixed; if accuracy does not fall, the second constraint is not doing the work the paper claims.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes NavEYE, a vision-centered AIS–radar–camera fusion pipeline for ISV situational awareness. MCGA associates AIS and radar via multi-feature Mahalanobis gating and Hungarian assignment; DAWF fuses dual-source tracks with distance-dependent error variances; TDSF stitches tracks under source loss via linear extrapolation and exponential decay; VARF associates fused geospatial tracks to ShipYOLO11n detections using joint normalized bearing and image-bottom distance. The authors release MAPFusion (135 min, three scenarios) with manual association labels and one GoPro-GPS reference track, report gains over JPDA/MHT/GRA, WAF/KF/EKF, and projection/bearing baselines, and describe a shipboard hardware/software system with vision and chart panels.

Significance. If the reported gains hold more broadly, the work is a solid systems contribution: a complete association–fusion–visualization stack on real shipborne multi-sensor data, with scenario splits, ablations on SOG/COG and bearing/distance, and an engineered NavEYE deployment. Strengths include real voyage collection, multi-object manual association labels, explicit sensor-error modeling for DAWF, and end-to-end identity/motion/appearance fusion that existing ECDIS-style systems often lack. The main scientific value is engineering integration and empirical validation rather than a new theoretical fusion principle; that is still useful for maritime perception if fusion accuracy claims are not over-generalized from a single reference track.

major comments (3)
  1. §4.1.2 and Table 3: AIS–radar fusion accuracy (DTW/RMSE/MAE for location, SOG, COG) is evaluated almost entirely against one cooperative ship (Zhoudou 16, MMSI 412439450) with a GoPro Hero 6 GPS logger on a fixed, regular route. Association F1 and VARF Acc use multi-object labels, but the headline claim of more accurate continuous fused tracks rests on this single trajectory. Please either (i) add multi-object/multi-route fusion GT (e.g., additional instrumented ships, high-grade RTK, or carefully validated multi-track references), or (ii) clearly restrict the fusion-accuracy claim to this reference case and avoid system-level generalization to dense/maneuvering/non-cooperative regimes where DAWF/TDSF are most needed.
  2. §3.2.2 and Table 4: TDSF is assessed only under simulated Missing AIS / Missing radar with PJ distance and APD versus hard switching (DS). There is no real dual-to-single source switching evaluation on MAPFusion with continuous GT, nor comparison to standard track-management alternatives (coasting, IMM, process-noise inflation). Given that trajectory continuity under observation-mode switching is a central contribution, real missing-source segments and stronger baselines are needed before claiming robust continuity for ISVs.
  3. §3.3.3 Eq. (32) and §6: VARF’s second constraint is the normalized vertical position of the detection-box bottom as a range proxy. The authors acknowledge pitch, waves, occlusion, and scale as limitations, but dense-scene Acc (narrow channel 74.89%, port 80.37% in Table 6) is load-bearing for the vision-centered claim. Please quantify sensitivity of Acc_association to camera pitch, sea state, and distant small objects (or ablate distance weight ω_d under those conditions) so readers can judge when the joint cost in Eq. (35) remains reliable.
minor comments (5)
  1. Free parameters (σ_lat/lon/sog/cog, g, β, λ, scenario-specific ω_φ/ω_d/γ) are listed in §4.2 without sensitivity analysis; a short sweep or justification would strengthen reproducibility.
  2. Table 6 header typo: “earing” should be “Bearing”; abstract/intro also have minor spacing/typo issues (e.g., “differentsensors”, “identificationsystem”).
  3. Fig. 7 COG jumps at 0°/360° are noted in text; plotting unwrapped COG or marking wrap events would make qualitative comparison clearer.
  4. ShipYOLO11n uses 1920×1080 while other detectors use 640×640 (Table 5); state more explicitly that the FPS/accuracy comparison is not resolution-matched.
  5. Related work could more clearly separate maritime AIS–radar track fusion from automotive radar–camera methods to avoid overstating novelty of the overall fusion idea versus the specific DAWF/TDSF/VARF design choices.

Circularity Check

0 steps flagged

No circular derivation: algorithmic fusion methods evaluated against external manual labels and independent GPS logger tracks.

full rationale

NavEYE is an engineering multi-sensor fusion paper. MCGA (Mahalanobis gating + Hungarian on lat/lon/SOG/COG), DAWF (inverse-variance weights from stated AIS GNSS and distance-dependent radar error models), TDSF (exponential time-decay stitching of extrapolated fused state with single-source observations), and VARF (joint normalized bearing/distance cost + Hungarian) are constructive algorithms, not claims that a quantity is derived from first principles and then shown equal to a fitted input. Evaluation uses external ground truth: manual AIS–radar and vision–AIS–radar association labels, and a GoPro Hero 6 GPS logger trajectory for fusion error (DTW/MAE/RMSE). Free parameters (σ, g, λ, scenario-specific ω_φ/ω_d/γ) are set by the authors and reported; they are not re-labeled as predictions of the same quantities used to fit them. Self-citations (prior Liu et al. maritime work; WUTDet for ShipYOLO11n fine-tuning) supply background datasets or related systems and do not import a uniqueness theorem or ansatz that forces the headline F1/DTW/Acc results. No self-definitional loop, fitted-input-as-prediction, or renaming of a known result as a derivation was found.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 4 invented entities

The central empirical claims rest on standard multi-target tracking math plus domain sensor-error models and several hand-chosen scalars (gating thresholds, decay rate, feature weights, radar distance coefficient). No new physical entities; invented items are engineering artifacts (methods, detector, dataset, system).

free parameters (6)
  • Mahalanobis residual stds (σ_lat, σ_lon, σ_sog, σ_cog) = 0.005, 0.005, 3.0, 3.0
    Set to 0.005, 0.005, 3.0, 3.0 for association cost; control which pairs pass the gate.
  • Association gate g = 3.0
    Threshold on Mahalanobis distance for candidate pairs (Eq. 4).
  • Radar distance error coefficient β and base variance = σ_base=8.1e-5; β free
    Define σ_r²(d) = 8.1e-5 + β d² used for DAWF weights; β is model coefficient not independently measured in the paper.
  • AIS/GNSS position variance σ_gnss² = 1.0e-8
    Fixed at 1.0e-8 for DAWF weight computation.
  • Timeliness decay constant λ = 0.5
    Controls exponential weight of historical fused state in TDSF (Eqs. 15–16).
  • VARF feature weights ω_φ, ω_d and cost gate γ = e.g. open: 0.65/0.35/0.85
    Scenario-specific hand settings (open/narrow/port) that directly shape association accuracy.
axioms (6)
  • domain assumption Mahalanobis distance with diagonal residual covariance is an adequate similarity for AIS–radar state matching.
    Used in MCGA Eqs. 1–4; assumes independent Gaussian-like residuals after unit scaling.
  • domain assumption Radar position error variance grows quadratically with range while AIS/GNSS error is approximately constant.
    Foundation of DAWF weights (Eqs. 5–8); motivated by beamwidth/cross-range resolution but not re-derived from radar hardware specs in the paper.
  • ad hoc to paper After dual-source loss, linear extrapolation of the last two fused states plus exponential time decay yields a useful prior for smooth stitching.
    TDSF Eqs. 12–19; motion model is constant-velocity style without process noise modeling.
  • domain assumption Image-centerline relative bearing from a bow-fixed, distortion-corrected camera equals own-ship relative bearing after gyro COG subtraction.
    VARF bearing path Eqs. 25–27; assumes installation alignment and FOV model.
  • ad hoc to paper Normalized vertical position of the box bottom is a usable proxy for relative range on the water surface.
    Eq. 32; paper later notes pitch/wave/scale sensitivity as a limitation.
  • standard math Hungarian algorithm on gated costs yields the intended one-to-one associations for evaluation.
    Used for both AIS–radar and vision–AIS–radar assignment.
invented entities (4)
  • MCGA / DAWF / TDSF / VARF method suite no independent evidence
    purpose: Named association and fusion procedures that implement the pipeline.
    Engineering constructs defined by the paper’s equations; evaluated only inside this work.
  • MAPFusion dataset no independent evidence
    purpose: Time-synchronized AIS/radar/video/GNSS/gyro labels for association and fusion benchmarks.
    New collection from one Zhoushan-area voyage; not shown to be public.
  • ShipYOLO11n no independent evidence
    purpose: YOLO11n fine-tuned on WUTDet for ship detection at 1920×1080.
    Transfer-learning artifact; performance reported only on MAPFusion.
  • NavEYE shipboard system no independent evidence
    purpose: Hardware/software stack for live fusion visualization and risk cues.
    System integration claim; engineering product of the methods.

reviewed 2026-07-11 · how reviews work

0 comments
Cite this review

Pith. "Pith review of NavEYE: Vision-Centered Multi-Sensor Fusion-Based Situational Awareness System for Intelligent Surface Vehicles." pith.science (2026). https://pith.science/paper/3H3HBLMP

@misc{pith2026260703915,
  author       = {Pith},
  title        = {Pith review of: NavEYE: Vision-Centered Multi-Sensor Fusion-Based Situational Awareness System for Intelligent Surface Vehicles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3H3HBLMP}},
  note         = {Machine review of arXiv:2607.03915}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

With the rapid development of sensor and artificial intelligence (AI) technologies, intelligent surface vehicles (ISVs) have gained increasing attention from academia and industry. Their intelligence, reliability, and safety depend heavily on situational awareness in complex navigational environments. To achieve high-quality perception, we develop a vision-centered multi-sensor fusion system, named NavEYE, by exploiting complementary sensors, including the automatic identification system (AIS), radar, and RGB camera. Specifically, we first propose a multi-constrained gated data association method (MCGA) to accurately match low-temporal-resolution AIS data with high-temporal-resolution radar data. Their fusion result is then obtained by selectively implementing distance-aware adaptively weighted fusion (DAWF) and timeliness decay-based stitching fusion (TDSF), which reduce the uncertainty caused by AIS or radar data loss in real-world sensing scenarios. Based on accurate and robust visual object detection, we further associate and fuse AIS, radar, and visual data through joint constraints of normalized bearing and distance features. According to the fusion results, comprehensive information related to ships of interest can be automatically obtained, helping enhance situational awareness and reduce collision risk for ISVs. The feasibility, robustness, usability, and effectiveness of the proposed multi-sensor fusion method and situational awareness system are demonstrated through extensive experiments on a real-world sensing dataset collected from AIS, radar, and camera. The experimental results show the superior performance of our fusion method in both quantitative and qualitative evaluations. In addition, the shipboard NavEYE system can promote navigational safety for ISVs in complex and dynamic environments.

Figures

Figures reproduced from arXiv: 2607.03915 by Haoyu Wang, Junxiong Liang, Mengwei Bao, Ryan Wen Liu.

Figure 1
Figure 1. Figure 1: The framework of the proposed vision-centered multi-sensor data fusion. Gyro denotes the gyrocompass device. MCGA, DAWF/TDSF, and VARF denote the proposed AIS–radar data association, AIS–radar data fusion, and vision–AIS–radar data fusion methods, respectively. The selection strategy is used to select the appropriate fusion method according to the available observation sources. Fused AIS–radar denotes the … view at source ↗
Figure 2
Figure 2. Figure 2: Adaptive AIS-radar data fusion strategy. DAWF is used when both AIS and radar data are available or when the observation state changes from single-source to dual-source; the available sensor trajectory is directly output under continuous single-source observations; and TDSF is applied when the observation state changes from dual-source to single-source. and 𝑀𝑟 denote the numbers of AIS objects and radar ob… view at source ↗
Figure 3
Figure 3. Figure 3: Timeliness decay-based stitching fusion method. It method can alleviate the problem of abrupt trajectory changes. 𝜂𝑓 , 𝜂𝑠 , and 𝜂𝑐 denote the extrapolated historical AIS–radar fused state, the current single-source observation state, and the stitched fusion result, respectively. 𝑤 𝑖 𝑠 (Δ𝑡) = 1 − 𝑤 𝑖 𝑑 (Δ𝑡) = 1 − 𝑒 −𝜆Δ𝑡 , (16) where 𝜆 is the decay constant that controls the weight decay rate. Finally, the s… view at source ↗
Figure 4
Figure 4. Figure 4: The illustration of the proposed vision–AIS–radar data fusion method. First, ShipYOLO11n trained via transfer learning extracts image-space and geospatial object sets from image data and AIS–radar fusion data, respectively. Then, bearing and distance feature deviation costs are computed through normalized and further fused to construct the association cost. Finally, the association cost matrix is generated… view at source ↗
Figure 5
Figure 5. Figure 5: The examples of MAPFusion dataset annotations. From top left to bottom right: (a) object bounding boxes in video frames, (b) ground truth AIS-radar data association, (c) ground truth AIS-radar fused trajectory, and (d) ground truth vision-AIS-radar data fusion. AIS, radar and vision sensors can have missed detec￾tions, false detections and inconsistent object numbers in real observations. Direct global ass… view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of AIS–radar data association results for different methods. MCGA (ours) exhibits fewer false associations across all scenarios, indicating better association stability and accuracy in complex maritime environments [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of AIS–radar data fusion results for different methods with both AIS and radar data available. 122.102 122.104 122.106 122.108 122.110 122.112 122.114 122.116 Longitude 29.797 29.798 29.799 29.800 29.801 29.802 29.803 29.804 29.805 29.806 29.807 Latitude Data loss (a) AIS data loss Ground truth AIS data Radar data TDSF(Ours) 122.102 122.104 122.106 122.108 122.110 122.112 122.114 122.116 Long… view at source ↗
Figure 8
Figure 8. Figure 8: Visualization of different AIS-radar data fusion methods under different simulated scenarios. and 𝑚𝐴𝑃50∶95 of 62.97%, outperforming the fine-tuned YOLO11n by 2.98% and 7.10%, respectively. Although its FPS decreases to 77.19, it is still much higher than the 1 Hz frame-sampling requirement of the NavEYE system. In the narrow channel and port/anchorage scenarios, many distant small objects are present [PIT… view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of results for different object detection methods. * indicates inference using pretrained weights on the COCO dataset. Methods without * indicate inference using weights fine-tuned on the WUTDet dataset. constraints achieves a precision of 62.94%. After introducing the SOG constraint, the precision of SCA increases to 72.68% (9.74%). This indicates that SOG information helps distinguish objec… view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of results for different vision–AIS–radar data association methods. Blue indicates correct associations, red indicates incorrect associations, and yellow indicates detected objects that are not successfully associated. crossing, parallel, and closely spaced objects. Across differ￾ent scenarios, the performance differences among methods are relatively small in the open water scenario. However… view at source ↗
Figure 11
Figure 11. Figure 11: Visualization of ablation results for vision-AIS-radar data association under different feature constraints and matching strategies. Blue indicates correct associations, red indicates incorrect associations, and yellow indicates detected objects that are not successfully associated. objects with similar bearings. This effect is particularly evident in high-density scenarios such as narrow channel and port… view at source ↗
Figure 12
Figure 12. Figure 12: The hardware architecture of NavEYE. AIS, radar, GNSS, and gyrocompass data are collected by onboard equipment, decoded by the data hub, and transmitted to the industrial computer through the network switch, while RGB camera video data are directly connected to the industrial computer via the network switch. Subsequently, the industrial computer performs multi-source data fusion. multi-source sensor obser… view at source ↗
Figure 13
Figure 13. Figure 13: Fused perception and anomaly monitoring in the NavEYE vision panel.Fusion perception and anomaly monitoring functions in the NavEYE vision panel. From left to right: (a) visualization of vision–AIS–radar fusion results, including information such as MMSI, longitude, latitude, SOG, and COG; (b) identification and warning display of vessels with collision risk. (a) Vision-AIS-radar fusion perception ercepti… view at source ↗
Figure 14
Figure 14. Figure 14: NavEYE position panel with layer control functionality. From top left to bottom right: (a) display of the own ship position, COG, and background map; (b) AIS data overlaid on (a); (c) radar data overlaid on (a); and (d) AIS–radar associated data overlaid on (a). data in list form, as well as GNSS and gyrocompass data in dictionary form. All these data are then input into the vision￾centered vision-AIS-rad… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 3 linked inside Pith

  1. [1]

    and Campari, R

    Sobolevsky, S. and Campari, R. , title =. Phys. Rev. E. , volume =. 2014 , pages =

    2014

  2. [2]

    and Wu, T

    Chen, Q. and Wu, T. T. and Fang, M. , title =. Physica A. , volume =. 2013 , pages =

    2013

  3. [3]

    arXiv preprint arXiv:2604.07759 , year =

    WUTDet: A 100K-Scale Ship Detection Dataset and Benchmarks with Dense Small Objects , author =. arXiv preprint arXiv:2604.07759 , year =

    Pith/arXiv arXiv

  4. [4]

    Sonar tracking of multiple targets using joint probabilistic data association , author =. IEEE J. Oceanic Eng. , volume =

  5. [5]

    IEEE Trans

    The interacting multiple model algorithm for systems with Markovian switching coefficients , author =. IEEE Trans. Autom. Control , volume =

  6. [6]

    IEEE Trans

    An algorithm for tracking multiple targets , author=. IEEE Trans. Autom. Control , volume=. 2003 , publisher=

    2003

  7. [7]

    Association and Fusion of Ship AIS and Radar Track Data , author=. J. Wuhan Univ. Technol , volume=

  8. [8]

    Ocean Eng

    Multi-target association algorithm of AIS-radar tracks using graph matching-based deep neural network , author =. Ocean Eng. , volume =

  9. [9]

    Ocean Eng

    Intelligent marine area supervision based on AIS and radar fusion , author =. Ocean Eng. , volume =

  10. [10]

    Radar and Automatic Identification System track fusion in an Electronic Chart Display and Information System , author =. J. Navig. , volume =

  11. [11]

    BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation , author =. Proc. ICRA , pages =. 2023 , publisher =

    2023

  12. [12]

    CenterFusion: Center-Based Radar and Camera Fusion for 3D Object Detection , author =. Proc. CVPR , pages =

  13. [13]

    Fusion of Camera-Based Vessel Detection and AIS for Maritime Surveillance , author =. Proc. ICAC , pages =. 2021 , publisher =

    2021

  14. [14]

    Image and AIS Data Fusion Technique for Maritime Computer Vision Applications , author =. Proc. WACV , pages =

  15. [15]

    Sensors , volume =

    Surface Vessels Detection and Tracking Method and Datasets with Multi-Source Data Fusion in Real-World Complex Scenarios , author =. Sensors , volume =

  16. [16]

    Autonomous collision detection and avoidance for ARAGON USV: Development and field tests , author =. J. Field Rob. , volume =

  17. [17]

    arXiv preprint arXiv:1702.00754 , year =

    Maritime situational awareness using adaptive multi-sensor management under hazy conditions , author =. arXiv preprint arXiv:1702.00754 , year =

    Pith/arXiv arXiv

  18. [18]

    PLOS ONE , volume =

    Research on shipborne aided navigation system based on enhanced traffic environment perception , author =. PLOS ONE , volume =

  19. [19]

    Review of research progress of autonomous and unmanned shipping , author =. J. Mar. Eng. Technol. , year =

  20. [20]

    IEEE Trans

    Sensors and AI Techniques for Situational Awareness in Autonomous Ships: A Review , author =. IEEE Trans. Intell. Transp. Syst. , volume =

  21. [21]

    Ocean Eng

    Data association for autonomous ships based on virtual simulation environment , author =. Ocean Eng. , volume =

  22. [22]

    Automatic Identification System AIS: Data Reliability and Human Error Implications , author =. J. Navig. , volume =

  23. [23]

    Sensors , volume =

    Analysis of Small Sea-Surface Targets Detection Performance According to Airborne Radar Parameters in Abnormal Weather Environments , author =. Sensors , volume =

  24. [24]

    IEEE Trans

    Asynchronous Trajectory Matching-Based Multimodal Maritime Data Fusion for Vessel Traffic Surveillance in Inland Waterways , author =. IEEE Trans. Intell. Transp. Syst. , volume =

  25. [25]

    Marine Vision-Based Situational Awareness Using Discriminative Deep Learning: A Survey , author =. J. Mar. Sci. Eng. , volume =

  26. [26]

    Sky and telescope , volume=

    Virtues of the Haversine , author=. Sky and telescope , volume=

  27. [27]

    Control problems of grey systems , author=. Syst. Control Lett. , volume=. 1982 , publisher=

    1982

  28. [28]

    Advances in neural information processing systems , volume=

    Faster r-cnn: Towards real-time object detection with region proposal networks , author=. Advances in neural information processing systems , volume=

  29. [29]

    European conference on computer vision , pages=

    Yolov9: Learning what you want to learn using programmable gradient information , author=. European conference on computer vision , pages=. 2024 , organization=

    2024

  30. [30]

    Advances in neural information processing systems , volume=

    Yolov10: Real-time end-to-end object detection , author=. Advances in neural information processing systems , volume=

  31. [31]

    Ultralytics YOLO , url =

    Jocher, Glenn and Qiu, Jing and Chaurasia, Ayush , license =. Ultralytics YOLO , url =

  32. [32]

    IEEE Internet Things Mag

    Multi-Sensor Data Fusion Meets Edge Computing for Intelligent Surface Vehicles , author=. IEEE Internet Things Mag. , volume=. 2025 , publisher=

    2025

  33. [33]

    IEEE Trans

    Graph learning-driven multi-vessel association: fusing multimodal data for maritime intelligence , author=. IEEE Trans. Intell. Transp. Syst. , year=

  34. [34]

    IEEE Trans

    Intelligent edge-enabled efficient multi-source data fusion for autonomous surface vehicles in maritime internet of things , author=. IEEE Trans. Green Commun. Networking , volume=. 2022 , publisher=

    2022

  35. [35]

    Ocean Eng

    Real-time vessel geolocation and automatic identification via video-AIS data fusion , author=. Ocean Eng. , volume=. 2025 , publisher=

    2025

  36. [36]

    Ocean Eng

    TAD-YOLO: A lightweight nearshore ship detection method for small USVs in maritime trade surveillance , author=. Ocean Eng. , volume=. 2025 , publisher=

    2025

  37. [37]

    Ocean Eng

    Orientation-aware dynamic vision-AIS association for UAV-based maritime surveillance , author=. Ocean Eng. , volume=. 2026 , publisher=

    2026

  38. [38]

    Autonomous Transportation Research , volume=

    Challenges for the Development of Maritime Autonomous Surface Ships , author=. Autonomous Transportation Research , volume=. 2026 , publisher=

    2026

  39. [39]

    Autonomous Transportation Research , volume=

    Concepts, Key Technologies, Applications and Development Trends in Autonomous Transportation Systems , author=. Autonomous Transportation Research , volume=. 2025 , publisher=

    2025

  40. [40]

    IEEE Trans

    AiOENet: All-in-one low-visibility enhancement to improve visual perception for intelligent marine vehicles under severe weather conditions , author=. IEEE Trans. Intell. Veh. , volume=. 2023 , publisher=

    2023

  41. [41]

    IEEE Trans

    Real-time multi-scene visibility enhancement for promoting navigational safety of vessels under complex weather conditions , author=. IEEE Trans. Intell. Transp. Syst. , volume=. 2024 , publisher=

    2024

  42. [42]

    Ocean Eng

    An enhanced CNN-enabled learning method for promoting ship detection in maritime surveillance system , author=. Ocean Eng. , volume=. 2021 , publisher=

    2021

  43. [43]

    Signal and Data Processing of Small Targets 2012 , volume=

    Measurement level AIS/radar fusion for maritime surveillance , author=. Signal and Data Processing of Small Targets 2012 , volume=. 2012 , organization=

    2012

  44. [44]

    The Journal of Navigation , volume=

    A study on AIS positional error analysis and transmission frequency requirements of attitude data for future vessel monitoring , author=. The Journal of Navigation , volume=. 2024 , publisher=

    2024

  45. [45]

    IEEE Transactions on Aerospace and Electronic Systems , volume=

    Coordinate conversion and tracking for very long range radars , author=. IEEE Transactions on Aerospace and Electronic Systems , volume=. 2009 , publisher=

    2009

  46. [46]

    2020 IEEE/ION Position, Location and Navigation Symposium (PLANS) , pages=

    Machine learning-assisted anomaly detection in maritime navigation using AIS data , author=. 2020 IEEE/ION Position, Location and Navigation Symposium (PLANS) , pages=

    2020

  47. [47]

    Journal of Fluids Engineering , volume=

    A new approach to linear filtering and prediction problems , author=. Journal of Fluids Engineering , volume=

  48. [48]

    Advances in control systems , volume=

    Application of state-space methods to navigation problems , author=. Advances in control systems , volume=. 1966 , publisher=

    1966

  49. [49]

    arXiv preprint arXiv:2605.26524 , year=

    CmIVTP: Cross-modal Interaction-based Vessel Trajectory Prediction for Maritime Intelligence , author=. arXiv preprint arXiv:2605.26524 , year=

    Pith/arXiv arXiv

This paper was first reviewed by grok-4.5 on July 11, 2026.