AI tech takes just minutes to convert GoPro videos into 3D coral reef maps

A new AI-based system allows coral reefs to be 3D-mapped faster and easier than ever before. It could prove integral to saving threatened reefs, by letting scientists study them both onsite and at their computers in their labs.

Currently, in order to obtain a 3D point cloud map of a coral reef, specially trained divers have to move around a section of that reef taking hundreds of photos from many angles.

A considerable amount of processing power is required to subsequently combine those images into one three-dimensional map. Even then, because of the extensive effort and computer analysis required, the map usually only shows a relatively small part of the reef – perhaps a few dozen meters.

That’s where DeepReefMap comes in.

Developed by scientists at Switzerland’s EPFL research institute, the system allows just one amateur diver armed with a single consumer-grade action camera to map several hundred meters of reef at a time. In fact, the main limiting factors are the camera’s battery life and the diver’s air supply.

As the diver swims over the reef, they shoot video of it with their downward-facing camera, catching images of the corals from a continuously shifting perspective. That footage is later processed by the DeepReefMap software, which utilizes deep-learning algorithms to convert the video into one big 3D composite “photo” that can be viewed from all of the camera’s original viewpoints.

Importantly, the system is able to compensate for the dim lighting and suboptimal water clarity that are so often part of underwater photography. By contrast, conventional reef-mapping technologies require brightly lit high-resolution images. What’s more, DeepReefMap can also assess the health of corals, plus it can classify them by species based on their shape.

In field tests conducted in the Gulf of Aqaba (in the Red Sea), a diver took five minutes to shoot a 100-meter (328-ft) stretch of reef. It took only another five minutes for DeepReefMap to convert that footage into a point cloud map.

In cases where a larger area of reef needs to be mapped, multiple cameras can be mounted on a single rig which is carried by one diver. For the Red Sea experiments, the scientists attached six GoPro Hero 10 cameras to a lightweight frame made of PVC piping. The cameras were spaced about 1 meter apart (3.3 ft), with three pointing forward and three pointing backward.

All six were shooting at 1080p/30fps, which proved to be completely adequate for DeepReefMap.

“With this new system, anyone can play a part in mapping the world’s coral reefs,” said Samuel Gardaz of the Transnational Red Sea Center, who took part in the study. “It will really spur on research in this field by reducing the workload, the amount of equipment and logistics, and the IT-related costs.”

A paper on the research was recently published in the journal Methods in Ecology and Evolution.

Source: EPFL

Source of Article