Rapid mapping of 3D environments

In April 2017, in response to an industry-posed challenge, Dr Adrian Clark and I developed a system for capturing video of the entire 4π steradians. The problem was testing such a system without getting operators in the field of view and to truly explore a 3-dimensional environment. Dr Clark, an experienced computer vision researcher, and myself, a specialist in human-machine interaction and underwater research, proposed to use computer vision processing to reconstruct a 3D model of the environment rather than just capture video of it. Furthmore, we proposed circumventing the problems of making the prototype hover by building an underwater system to capture structures such as coral reefs and wrecks, using floats within it to achieve neutral buoyancy.

The rig was tested at a marine research site in Dominica in the Caribbean during July and August 2017. The 3D models reconstructed from the images captured from the rig show accuracy and clarity at both small and large scales. Work is currently under way to optimise the procedure and increase the speed of data collection as time spent in the field is the rarest commodity in this type of research; however, the current speed of environment mapping is far greater than any other visual method currently used.

The research is spawning new collaborations with interested researchers including Coral Reef Research Unit in Essex's School of Biological Sciences, where we are now working on several new reef conservation projects.

The University of Essex has been active in underwater robotics for some years, and is keen to extend existing work on robotic fish and stereo vision systems. To this end, the faculty has funded a PhD scholarship to develop this research.

Disagreement and Ambiguity in Language Interpretation (DALI)

Natural language expressions are supposed to be unambiguous in context. Yet more and more examples of use of expressions that are ambiguous in context are emerging. In previous work using crowdsourcing we demonstrated that ambiguity in anaphoric reference is ubiquitous, and there is frequent disagreements in annotation. Using the Phrase Detectives Game-With-A-Purpose to collect massive amounts of judgments online, we found that up to 30% of anaphoric expressions in our data are ambiguous. These findings raise a serious challenge for computational linguistics (CL), as assumptions about the existence of a single interpretation in context are built in the dominant methodology, that depends on a reliably annotated gold standard.

The goal of the DALI project is to tackle this fundamental issue of disagreements in interpretation by using computational methods for collecting and analysing such disagreements, some of which already exist but have never before been applied in linguistics on a large scale, some we will develop from scratch. Specifically, we will develop more advanced games-with-a-purpose to collect massive amounts of data about anaphora from people playing a game.

Groupsourcing: Identifying wildlife on social media

Hundreds of thousands of images are being posted to social websites, showing a unique glimpse of the underwater world. Analysis of a small subset of these images show very high accuracy of image tagging (93% were annotated correctly) which makes them useful as a primary data source for conservation research. By analysing the text associated with the uploaded images it is possible to map where different species live around the world, what they eat, what eats them and whether changes are occurring to their populations.

Social networks can be seen as decentralised and self-organised crowdsourcing systems that are becoming increasingly popular. Tasks are created by the users, so they are motivated to participate, and the natural language of the interface allows them to express their emotions whilst solving the tasks.

We call this approach groupsourcing, completing a task using a group of intrinsically motivated people of varying expertise connected through a social network.