Automating Metadata Logging Through Artificial Intelligence

In 2007, NASCAR designed and implemented a media asset management solution for their sport. Now, in 2018, the library has grown to more than 500,000 hr of content containing video, audio, and still images dating back to 1933—one of the most vast sports libraries in the world. Since the inception of the library, NASCAR has employed a staff to manually apply metadata to each video frame, amassing 10 million entries. While the logging by our staff has been impressive, the incoming data avalanche cannot be addressed by using the current system, let alone the glacier of data hiding in the archive. At present, we have calculated that it would take our existing staff nearly 150 years to log all of the historical content as it stands today. Over the past ten years, we have extensively analyzed open-source and proprietary tools aimed at dealing with the data logging gap and have determined that machine learning is the ideal solution to address metadata logging on large-scale media libraries. As machine learning has become more accessible through the scalability of cloud computing, training data, and implementing convolutional neural networks, it is now within the reach of media production companies and asset stakeholders. NASCAR is on the verge of revolutionizing how all data asset management systems can be restructured in the future to integrate machine learning to harness efficiencies in metadata logging.

Print ISSN

1545-0279

Electronic ISSN

2160-2492

Published

2019-10

Content type

Original Research

Keywords

Artificial intelligence (AI), convolutional neural networks (CNNs), machine learning, metadata tagging, tensorflow, transfer learning, video logging

DOI

10.5594/JMI.2019.2932163