Mussel Float Mapper

Overview

The Aquaculture industry in New Zealand contributes over $650 million per year to the economy with Green-lipped mussel farming contributing 30% of this. A key challenge for mussel farmers is the loss of mussel floats due to adverse weather conditions causing pollution and significant costs estimated to be $500k per year.

This project was a summer research project with the aim of mapping mussel float locations using video footage captured by vessels along with GPS data. This combines computer vision techniques to identify mussel floats in the vidoe footage, research into 3D reconstruction, image reprojection, and 3D game engine tooling exploration.

My contribution to this ongoing research area was the combination of various previous research, being shoreline detection, 3D reconstruction and object detection. Limitations were discovered with combining the existing research due to the unique challenge of estimating camera placement, and properties such as focal length and alignment of the camera.

I explored the use of Unity 3D as a tool to build a heightmap of the surrounding environment and simulation the camera placement and properties. This allowed for manual alignment of the camera to the generated 3D environment. Applying the object detection technique to perform raycasts onto an ocean surface mesh allowed for the generation of world coordinates for each detected mussel float.

This led gave very promising results, but due to the inaccuracy of GPS data of 2-metres. The results led to an overlap of mussel floats, but the project demonstrated the potential of this technique for future research.