Welcome to www.oceancv.com - the home of the open source library and tools for ocean image analysis. Merge requests to OceanCV are very welcome! OceanCV is C++ based and relies on the amazing OpenCV library.
OceanCV provides dedicated machine learning and high-throughput image analysis algorithms tuned for application to underwater images.
They were developed in international research projects and applied in detection, classification and quantification use-cases for benthic fauna, mineral resources and others.
Checkout the documentation for an overview of the class structure and the examples to get startet with using and expanding OceanCV.
You can use OceanCV as CPU-only, but that would be a massive waste of time as the GPU-enabled methods have shown a more than 1000x increase in analysis speed!
Not sure whether OceanCV is the right tool for you or are you looking for a partner to outsource your scientific image analysis tasks to?
Talk to us to evaluate consultancy options. We are also keen to collaborate in research projects depending on machine learning and images.
Laser points are a robust and widely used means of determining the scale of image data in meters. Knowing this scale for each individual photo or video frame is essential for applications that depend on object sizes like biomass assessments of object identification. The DeLPHI system, consisting of laser point detection methods for arbitrary laser point colours and training and application tools to run those algorithms, provides that functionality. It has been described here: https://doi.org/10.3389/fmars.2015.00020.
The DeLPHI algorithms are available in the OceanCV uwi module and the tools in the bin/uwi module.
Semantic annotations are required not only for manual image analysis but also for training machine learning algorithms. But creating proper annotation data requires thorough planning, conducting and quality-checking of annotations. The Recommendations for Marine Image Annotation can help with that.
The machine learning methods in OceanCV rely on such robust annotations. The recommendations inspired the efficiency-optimised annotation tools in the annotation software BIIGLE and were further developed into BIIGLE's "Largo" QA/QC tool.
Assessing the size of individual nodules and the seafloor coverage and total abundance of poly-metallic nodules across square kilometers requires analysing big image data sets. The fully-automated CoMoNoD algorithm can help with that. Running it delineates individual nodules in images and computes nodule distribution statistics and particle size metrics.
With these numbers, the natural resource distribution can be mapped. Also, nodule-associated fauna distribution patterns can be investigated. Monitoring of mining activities is enabled by comparing pre- and post-mining seafloor imagery to quantify the size and strength of the mining impact - e.g. in terms of a re-deposited sediment plume.
Computer vision for ocean images