ORIONS AI is an AI CITY service to create and train video algorithms. Built to take advantage of Orions cloud-based design, Orions AI provides elastic scalability to meet the requirements of any task, from basic algorithms driven by simple business rules, to complex smart vision systems that require the application of human and algorithmic intelligence to hundreds of thousands of video hours.
Algorithm Training Example – Baseball Pitches
A small team of human taggers reviews and tags pitches in a subset of video segments.
Orions AI scans the tagged video segments to identify changes in the scene, such as a single object moving at higher velocity than any other object in the scene.
Now the algorithm can quickly process hundreds of hours of baseball footage and identify the segments that likely have pitches based on the velocity/object detection.
Human taggers review a random sampling of both the newly identified pitch video segments as well as segments the algorithm determined to NOT include a pitch. Taggers identify segments with errors (false positives or false negatives) and feed these segments with the correct data back into the algorithm.
The machine learning cycle continues until Orions AI is achieving an acceptable level of accuracy at scale.