We Understand Motion Analytics
Predicting the location and behavior of data and drawing meaningful conclusions from their motion is conceptually and mathematically challenging. Many theoretical approaches have been developed, and we understand which techniques are best applied to a particular problem.
For example, consider a set of ships moving on the surface of the ocean. Embedded in the movement of these ships is a whole host of information about the behaviors and intentions of their owners, captains, and crews. Why does one vessel take a seemingly inefficient route between two places in perfectly good weather? Does it want to avoid being seen by other ships?
Why do two fishing vessels from different countries rendezvous on the open ocean? Are they illegally exchanging their catch, or engaged in human trafficking? Why does a cruise ship's movements closely follow the wind and ocean currents? Is it adrift and in distress?
Graph Theoretical and Network Techniques
Sometimes motion data is best understood as a collection of route and destination (edges and vertices). Graph theory is a rich mathematical discipline with established solutions to many common problems such as connectedness, pathfinding, distances and probabilistic prediction. Interconnectedness in human relationships (sometimes called social network analysis, but not just applied to application like Facebook and Twitter) can also be explored effectively through graph theoretical techniques.
Geometric and Pattern Matching Techniques
Sometimes motion is expressed in regular, predictable or distinctive mathematical or geometric patterns. A fishing trawler might leave port, cast its nets in a particular manner, retrieve them at a later time and return to its home port whereas a ferry may travel back and forth between two destination in a straight line, never deviating from its pattern. When drawn out in time series or other projections, these ships create unique geometric patterns. Shape and object recognition techniques, series pattern detection and genetic algorithms help us understand, organize and classify this kind of motion.
Machine Learning Techniques
Machine learning techniques help make sense of complex, unorganized and seemingly random data sets. Clustering identifies data elements that share a common set of features or properties. Correlation tells us whether some features of data (e.g. the size or class of ship) help us predict other features (such as its speed or location). Dimensionality reduction helps us reduce the data search space to focus on only those piece of information that matter. The most powerful techniques, prediction techniques, tell us what an object in motion should do based on past behavior as well as whether it is behaving outside the norm.
Our Work in the News
Google, Partners Target Illegal Fishing with New Technology
"Google Inc. has teamed up with mapping company SkyTruth and marine advocacy group Oceana to create a new tool aimed at reining in illegal fishing worldwide... Read More"
The Plan to Map Illegal Fishing From Space
"ILLICIT FISHING GOES on every day at an industrial scale. But large commercial fishers are about to get a new set of overseers: conversationists- and soon the general public-armed with space-based reconnaissance of the global fleet... Read More"