Adam Motaouakkil, November 26, 2025
Definition
Video Analytics is a field where videos are analyzed through statistical models to extract data. That data is then used in cases such as surveillance and retail analytics.1 Anonymous Video Analytics (AVA) is the use of anonymized video for analytics. It gathers information on relevant data such as the presence of an individual and does not record identifiable characteristics such as facial features and body pose.2
Digital Signage encompasses all signs displayed publicly through electronic means. The technology is also called Digital Out of Home signage (DOOH). LED panels that display advertising serviced by servers and software are part of this field.1
Sensors
A typical system comprises cameras and may involve Wi-Fi trackers or other systems that pick up on data-sharing agreements users sometimes opt into such as free Wi-Fi.3 Cameras are used because they are suitable to identify gender, estimate age, and gather other types of data that can be inferred by machine learning models. There are also other trackers that are used for detecting presence and do not record images or video.1
System Architecture
Typically, the camera feed is fed to a data analytics service. This can be done on edge devices that process the data in the Content Management System (CMS).4 The analytics software is usually run on the cloud and returns relevant data. For Anonymous Video Analytics, gender and age are returned. The accuracy of those results are vendor specific. The data is produced by running images through machine learning models that can classify input frames. Models are “taught” to recognize input and then used in commercial systems as they are assumed to have been trained well enough to recognize images accurately.5 Classifications such as heteronormative gender identity are more accurate because of physical appearance. Age, however, is trickier as physical appearance does not predict exact age. Someone can look old, but they might be 50 or 70.
Popular image classification machine learning models such as YOLO are served through platforms that are easy to integrate. The models can be cheap to run and to deploy on a server system.6
In a parallel process, the ad-servicing software runs through the cloud and services edge devices. The software provides buyers with permanent and flash ad availability. They can supply data on location, demographics, duration of ad, and ad rules to ensure compliance with the rest of companies’ ads. Through an automated process, companies bid for the ads and when accepted, send their media to be displayed.7
There is no set standard for bidding strategies. Algorithms seek to maximize certain metrics such as retail success and select the company that best fits the criteria.8
Important Notes
The Digital Signage industry is rife with vendor consortiums, proprietary standards, and technology that is not open to the public. By nature of the field, its computer systems have also evolved behind secretive Intellectual Property except for slivers of attempts at opening up standards.1
The data that machine learning systems are trained on are labelled by people who are required to adhere to set standards and expectations that do not always involve under-represented groups. Vetted datasets are not immune to human biases and prejudices of all forms due to excluding those groups from consultation.9 Data labelling is an important industry for companies that provide Large Language Models. Data labellers are frequently exploited by data annotation operations to meet the industry demands for more datasets.10
References
1. Yokinobu Taniguchi, “Content Scheduling and Adaptation for Networked and Context-Aware Digital Signage: A Literature Survey” ITE Transactions on Media Technology and Applications (2018): 18-29. https://www.jstage.jst.go.jp/article/mta/6/1/6_18/_pdf/-char/en
2. Zhang, Shikun; Feng, Yuanyuan; Bauer, Lujo; Cranor, Lorrie Faith; Das, Anupam; Sadeh, Norman, “Did you know this camera tracks your mood?: Understanding Privacy Expectations and Preferences in the Age of Video Analytics”, Proceedings on Privacy Enhancing Technologies (2021): 282-304.
https://users.ece.cmu.edu/~lbauer/papers/2021/popets2021-video-prefs.pdf
3. Brian T. Horowitz, “5 Things You Didn’t Know Free Wi-Fi Can Do for Your Business”, PC Mag (2019).
https://www.pcmag.com/news/5-things-you-didnt-know-free-wi-fi-can-do-for-your-business
4. Osaka Tomoyuki, “Trends in Digital Signage Solutions”, NEC (2011): Vol. 6 No. 3.
https://www.nec.com/en/global/techrep/journal/g11/n03/g1103pa.html
5. Abirami Vina, “The evolution of object detection and Ultralytics’ YOLO models”, Ultralytics (2024).
https://www.ultralytics.com/blog/the-evolution-of-object-detection-and-ultralytics-yolo-models
6. Ultralytics Code Repository, maintainers of YOLO https://github.com/ultralytics/ultralytics
7. “Buying & Selling: How do digital out of home transactions work?” Digital Out of Home, A Primer, Section 3, Interactive Advertising
Bureau (2019). https://www.iab.com/wp-content/uploads/2019/03/DOOHSection3.pdf
8. Alex Rogers, Esther David, Terry R. Payne and
Nicholas R. Jennings, “An Advanced Bidding Agent for Advertisement Selection on Public Displays”, 6th International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS 2007) (2007). https://www.researchgate.net/publication/221455224_An_Advanced_Bidding_Agent_for_Advertisement_Selection_on_Public_Displ
9. Datasets, Bias, Discrimination – University of Toronto Libraries Research Guides. Last updated: August 2025
https://guides.library.utoronto.ca/c.php?g=735513&p=5297043
10. Billy Perrigo, “Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic”, Time (2023). https://time.com/6247678/openai-chatgpt-kenya-workers/
