As a computer vision group, we work with video and image processing, mainly for underwater applications.
Initially, underwater surveys with unmanned underwater vehicles had their major tradeoffs with the technical specifications. Tradeoffs wich led to various flaws, such as low-quality video capturing and low degree of control over it. Today, one of the bottleneck links in the chain of surveys is a man behind the unmanned underwater vehicle (UUV), who operates it and trying to get an adequate picture of what's being viewed on the screen.
Low quality of the content can be explained by several facts: cameras are too cheap, too much mud or organic particles, not enough ceiling light.
We can propose a solution to the case, given the fact that research on parts of the problem has been made. Our turn is to make further research and to bring it to the customers polished, fast, and beautiful to make the work of UUV operators simpler and surveys cheaper.
To achieve it, we are developing our approach, which includes several traditional methods of processing the video with the usage of modern architectures based on deep neural networks.