360º Video Object & Scene Detection

Research

Advertisers want to know what viewers are looking at in 360º videos when part of a virtual reality experience. When users look in all directions, tracking becomes more difficult. How do you tell how many times a brand or product is viewed? Heat-maps provide coordinates, but machine learning is needed to understand the entire scene and the objects in it. Image processing allows for a greater understanding of user experience, while saving time and money by quickly and accurately building qualitative data sets. Our research discovered the best methodologies for breaking equirectangular frames into cube maps and using Caffe’s trained machine learning libraries to identify objects in a scene.

Elan Colello | 360º Video Object & Scene Detection