The proliferation of digital video media provides the opportunity for expansive analyses on widely available, commonly used content. The structure of video, though, is often complex, presenting a barrier to automated segmentation which could facilitate personalization and analysis. Effective segmentation of videos into relevant episodes requires advancement in computer programming tools to accurately and reproducibly demarcate video files based on broadly applicable criteria. Previous approaches to video segmentation rely heavily on transitions between shots; however, broad application of this approach is limited because the demarcation of a shot does not necessarily correspond to a transition that is interesting or relevant to viewers. This technology describes a computer programming method based on advanced statistical techniques to enable video segmentation of structurally complex content. The technology could be readily applied to a variety of video-based media, such as sports broadcasts, to facilitate editing, analysis, and video-clip browsing.
Advanced statistical techniques enable automated segmentation of video files containing structurally complex content. High-level contextual cues from the video are utilized to perform segmentation without relying on shot transitions. This approach is particularly advantageous for videos of unpredictable content and could reduce size by eliminating extraneous material.
This computer algorithm has successfully segmented a video of a soccer game into moments of play and break in game action, demonstrating the utility of this method for segmentation of complex videos of commonly watched content.
Patent Issued (US 6,865,226
Tech Ventures Reference: IR M02-070