A Computer Vision-Based Method for Feature Extraction From Ocean Wave Images and Sea Surface Elevation Inversion

Stereo photography is a mainstream non-contact technique for wave field measurement, yet its accuracy calibration typically relies on contact-based devices such as buoys, which suffer from difficulties in deployment and recovery, vulnerability to harsh sea states, and limited spatial coverage. To address these issues, this paper integrates monocular machine vision with conventional image processing techniques to construct a lightweight model for wave image feature extraction and sea surface elevation inversion. The model employs a blue-channel separation strategy based on the optical attenuation characteristics of seawater, utilizes adaptive threshold segmentation and morphological cleaning to achieve robust wave contour detection, and establishes a triangular geometric height measurement model to suppress measurement errors caused by image tilt. Furthermore, based on the camera imaging geometric model, the conversion relationship between pixel dimensions and actual distances is derived to accomplish sea surface elevation inversion. A parameter-tunable virtual ocean simulation platform is adopted to generate dynamic wave fields, and the instantaneous elevation data output from the platform serve as the simulation ground truth for quantitative accuracy assessment. The results indicate that the relative errors between the model-inverted elevations and the simulated true values range from 2.89% to 12.50%, with an average relative error of 6.84%, which verifies the feasibility and effectiveness of the proposed method in elevation inversion. This approach provides a low-cost technical pathway for wave field observation with reduced dependence on field calibration equipment.