BENCHMARK FOR ANONYMOUS VIDEO ANALYTICS
Welcome to DPAA & Intel’s Benchmark for Anonymous Video Analytics
for Digital Out-of-Home audience measurement.
Anonymous Video Analytics (AVA) is a camera technology that enables
real-time understanding of audiences exposed to digital out-of-home advertisements
ABOUT THE BENCHMARK
Understanding the number of people exposed to digital screens in the real world is important to help advertisers measure their return on investment. However, while the adoption of Anonymous Video Analytics (AVA) are increasing, no commonly accepted benchmark exists to evaluate its performance. In this paper in partnership with INTEL and Queen Mary University of London (QMUL), we propose the first AVA benchmark for digital out-of-home audience measurement ; one that evaluates audience counting, and audience demographics.
AVA ensures the preservation of the privacy of audience members by performing inferences and aggregating them directly on edge systems, without recording or streaming raw data. AVA relies on person detectors or trackers to localize people and to enable the aggregate estimation of audience attributes, such as their demographics.
Strategically leveraging the intelligence of computer vision for audience measurement elevated URW’s data-centric media offering. Our ability to share creative-specific demographics and viewability analytics, in real-time, allows brand partners to optimize active campaigns and ensure maximum ad effectiveness. The live intelligence can also be utilized to accurately and authentically validate an impression, based on individuals who are detected within 20 feet of a screen while content is playing.
URW was the first in the US DOOH industry to offer an impression-based guarantee based on computer vision’s accuracy and campaign transparency. The ability to offer brands a live, location-based and eyes-on on advertising platform with in-depth campaign intelligence is unparalleled in today’s DOOH marketplace.
Although Anonymous Video Analytics have been in use for over a decade, as an industry we have long suffered from the lack of a precise and objective methodology that can be used to validate the metrics we provide. QMUL and Intel, with their excellent work, have fulfilled a longstanding need and paved the way for greater accountability, helping the adoption of AVA and fostering even more exciting innovation in the world of DOOH.
Our aspiration to innovate in the AI space has been hugely enabled by technology partners like Intel.
By successfully creating the benchmark for testing the accuracy of computer vision based products, Intel has allowed us to build, test, and validate new use cases rapidly. This has driven us to improve accuracy and performance in anonymous measurement especially for digital out of home.
Using OpenVino and its toolkits, we are able to train and optimize machine learning models faster and cost effectively. This and coupled with our ability to use synthetic data to train ML models has significantly improved our ability to consistently innovate and provide the highest quality of service to our customers.
KEY TAKEAWAYS
- Tracking algorithms, which track a person in the short-term for a duration of few frames, reduce redundancies and are more effective than just the detection algorithms in providing a more accurate result for people counting
- Higher fps does not always equate to higher accuracy especially for Audience analytics. Balanced fps (6-15) with tracking algorithm, as described above, provide the most optimal results
- Median performance is impacted equally in the cases where there is partial or heavy occlusion (higher number of people, obstructions etc.) irrespective of whether the workload is run on a CPU or GPU
AVA for DOOH
Aims to quantify people count for opportunity to see (OTS)
AVA for DOOH
Aims to quantify attributes for individuals with an opportunity to see (OTS)
Click Here to view the Benchmark Website
Provided by the Centre for Intelligent Sensing
Click Here to view the Code Repository
Provided by Queen Mary University of London
DISCLAIMER:
Intel is committed to respecting human rights and avoiding complicity in human rights abuses. See Intel’s Global Human Rights Principles. Intel’s products and software are intended only to be used in applications that do not cause or contribute to a violation of an internationally recognized human right.