Examples from use cases
Using the Skarv Autonomy Engine (SKATE) we are able to autonomously map the seafloor and allow for adaptive optimization schemes that can increase the amount and relevance of the data being collected. The system automatically adjusts the speed and height to ensure that the quality of the data is maintained. Using Skarvs own stereo-camera system we are able to generate photo-mosaics with resolutions below 0.5cm (shown in the image to the left).
Automatic detection of known objects
Using custom trained artificial neural nets we are able to detect, segment, and track objects on the seafloor or on structures. The systems can be trained on new data to do custom detection of underwater objects such as debris, pipelines, anchors, archaeology, damage types, etc.
Anomaly detection of new objects
Finding unknown object without prior information is also possible using attention-based neural architectures. The system can be used to look for lost objects, damages, irregularities, and conspicuous elements.
3D reconstruction of underwater scenes
Using structure from motion (SfM) and CNN-based feature descriptors we are able to reconstruct 3D-information from difficult underwater scenes amidst schools of fish and suspended particles. Our tools can deliver geo-referenced 3D models in standard formats for integration into BIM-models.