Imagine how frustrating it can be for a retailer to spend thousands of dollars on a digital signage network and not know if the content it displays is effective in capturing and holding a viewer’s attention. Without solid data, it’s hard to judge advertising effectiveness and decide which advertisements should be played more and which need improvement. Equally difficult is determining business value.


In this post, we look at an example from an Intel white paper on an Intelligent Advertising Framework that describes setup recommendations for a solution that uses Intel® Audience Impression Metrics Suite (Intel® AIM Suite) and data mining techniques to collect, analyze, and put viewer data to work to improve advertising effectiveness. This particular solution could be used to automate the correlation of viewer demographic information and dwell (viewing) time with playlists—and even point-of-sale (POS) data— to help retailers more accurately measure the effectiveness of their advertising for business value. Equally important, and what I want to focus on here, is how the solution can apply machine learning to intelligently automate and continuously improve the selection and play in real time of the most effective ads to the current viewer in front of a sign.


Anonymous Viewer Analytics and Intel® AIM Suite

Intel AIM Suite is a software solution that uses pattern detection algorithms called Anonymous Viewer Analytics (AVA) to scan real time video feeds. These algorithms look for facial patterns to determine gender, age bracket, and dwell time. To ensure privacy, nothing in the process recognizes or records the individuals who passed in front of the cameras. The data is logged and then the video destroyed on the fly.


The data generated by Intel AIM Suite gives marketers and business communicators valuable new insights into the numbers and characteristics of the audience, including how long an individual looked at the sign. By reporting on what actually happened within the proximity of displays, Intel AIM Suite helps communicators improve on what they’re doing.


Intel AIM Suite is available from a wide range of sources, including members of the Intel® Intelligent Systems Alliance. For instance, Bsquare offers it as an embedded software image. Avnet can include it as part of their Rorke Global Solutions Paladin Media Player and Venture Corporation as part of their eIPC300 Embedded PC. It is also available from digital signage specialists AOPEN and Dynasign.


How Intel AIM Suite Works

Intel AIM Suite relies on camera sensors built into a display. The software is loaded on media players connected to these camera-equipped displays. Intel AIM Suite’s Audience Counter measures the numbers of consumers who enter and pass through a zone and have the opportunity to see sign messaging. The Audience Counter brackets the patterns of faces actually looking at the sign, anonymously logging these as events and capturing data such as what was observed and for how long (Figure 1).


Looking at sign viewing time image.jpg

Figure 1. Intel® AIM Suite Audience Counter anonymously tracks viewers.


The steady, real-time stream of captured patterns is processed by unique computing algorithms to detect the numbers, genders and age brackets of viewers in an area. Events are time-stamped and logged, and data is available in real-time for systems operating in parallel to respond. For digital signage, that could mean actions like playing back tailored content or offering samples or experiences tailored to detected viewer types.


Intel AIM Suite is managed remotely using the Intel AIM Manage website. Intel AIM Analytics are viewed through this website as well. Intel AIM Suite can also be integrated with a digital signage system using Intel® Retail Client Manager (Intel® RCM), a content management solution. This enables conditional play where specified content will play if a person fitting a selected gender/age profile is identified in front of a sign.


Integration Using Application Programming Interfaces (APIs)

In order to seamlessly access, work with, and merge data to develop highly detailed, insightful analytics and create innovative user experiences, Intel provides three different Intel AIM Suite APIs to third-party developers and providers (Figure 2).



Figure 2. Intel® AIM Suite APIs.


Developers can use these APIs can set up systems that help retailers extract data from AIM Suite to help calculate sell-through and conversion ratios for ROI determinations, linking the timing of messages and numbers of viewers to bottom-line impacts.


The three Intel AIM Suite APIs include:

·        Low-Level Socket-Based API. This API allows developers to connect to the Intel AIM Suite application using TCP/IP. This connection enables a platform-independent interface to Intel AIM Suite video analytics products and services, with support for both local and remote communication between AIM Suite and third-party applications. (See the Socket API Reference Manual for more information.)

·        High-Level Socket-Based API. This API helps paint an even broader picture of the viewing audience by providing information on total number of viewers and face-detected and face-lost events. It also provides details from the low-level API.

·        Web (HTTP) API. This API is used by a customer’s web server to pull information from the Intel-hosted AIM Analytics server. The Web API enables data extraction through a PHP script over HTTP. This returns information in a comma-separated values (CSV) format for each impression in a desired date range. The information can include the date, impression start time, viewing time, gender, and age bracket.


Intelligent Advertising Framework for Digital Signage

Using these APIs in a proof of concept, Intel developers designed an Intelligent Advertising Framework (IAF) based on using AVA technology and data mining concepts to achieve real-time advertising targeting. IAF can be used to measure and improve the advertising effectiveness (and thus ROI) of a digital sign.


The Intel IAF team primarily considered targeted advertising techniques that required the digital signs to have the capability to dynamically select and play advertisements for specific audience (based on gender and age bracket) according to the past effectiveness of these advertisements in capturing and holding the attention of past viewers sharing the same profile. They called this seeing-based targeting since the sign “sees” the audience through Intel AIM Suite and then posts appropriate ads.


To do this, they added a data mining module (DMM) to the Intel AIM Suite setup that could correlate the viewer events with player events (what ads were played at what time and in what location), as well as other context information (see Figure 3).


IAF setup.jpg

Figure 3. Simple diagram of the IAF architecture. “VA” stands for video analytics.


The DMM uses data mining algorithms to discover to what extent a specific audience is interested in a specific advertisement or type of content. The demo plays default advertisements in a predefined or random order when no audience is present. Once an individual is actively watching the display, the demo will notice, capture, analyze and track a viewer’s face in real time. The demo will then play content that the audience would be most interested in and offer recommended content based on predictive modeling (Figure 4).


IAF Test ad.JPG

Figure 4. This screen shot of the IAF demo interface shows the captured human face in a green circle. On the bottom left is a list of default advertisements, a list of recommended advertisements for the current audience, and a list of relevant advertisements to the current playing one. On the top left, is the current playing advertisement. Bottom right is a graph of audience count, in this case one young adult male.


In addition to watching the played advertisement, the audience can select and play any of the recommended advertisements. When the audience selects other advertisements, a new list relevant to the selection is provided. The viewer interaction is recorded as an event and used as input to the predictive modeling to discover associations and improve the accuracy of future recommendations for viewers fitting similar profiles.


The test compared three algorithms in the Microsoft SQL Server Analysis Service (SSAS): Decision

Tree, Association Rule, and Naïve Bayes. Results showed the highest targeting accuracy with the Decision Tree algorithm. The Intel team concluded that a system using IAF could eventually achieve a targeting accuracy of up to 90 percent if deployed in locations where there was considerable gender and age differentiation.


A much more detailed explanation of the IAF architecture and the rules for weighting what content to display can be found in the Intel white paper “Realizing Targeted Advertising in Digital Signage with AVA and Data Mining.”


Automating Real-Time Targeted Advertising

While Intel AIM Suite is an extremely effective tool for the manual setup of conditional play based on factors like gender and age, the IAF proof of concept demonstrates a way to automate this process through modern data mining techniques. Developers and system integrators could use such a solution to enable their retail customers to dramatically improve digital signage advertising results, continuously improving its ability and effectiveness in matching content to viewer.



Learn More

Contact Featured Members:


Solutions in this blog:

·        Bsquare Intelligent Systems for Retail

·        Intel® Audience Impression Metrics Suite (Intel® AIM Suite)

·        Intel® Retail Client Manager (Intel® RCM)

·        Rorke Global Solutions Paladin Media Player (Avnet)

·        Venture eIPC300 Embedded PC


Related topics:

·        Sensing and Analytics - Top Picks (blogs, white papers, and more)

·        Digital Signage - Top Picks (blogs, white papers, and more)



Avnet and Venture Corporation are Associate members of the Alliance. Bsquare is an Affiliate member of the Alliance. AOPEN and Dynasign are general members of the Alliance.


Mark Scantlebury

Roving Reporter (Intel Contractor), Intel® Intelligent Systems Alliance

Associate Editor, Embedded Innovator magazine