There are currently more than 1.2 billion Internet-enabled desktop, mobile and tablet devices equipped with camera sensors. Camera sensors are expected to double by 2017, as wearable devices, smart TVs and other devices incorporate new smart features and capabilities. 3 Million digital displays in the US are already web-enabled, reaching over 70% of US teens and adults every month in public venues. According to ABI Research, by 2020 the majority (60 percent) of the total installed base (50 billion units) of Internet of Things (IoT) devices will incorporate nodes/sensors.
Retailers and market researchers are beginning to use pattern detection technology to understand viewing audiences. The use of this technology enables retailers, packaged goods brands, agencies and operators of facilities such as malls, airports, colleges, and museums to better understand and communicate with their consumers.
What is Anonymous Video Analytics?
AVA analyzes millions of pixels per second and anonymously detects general traits of viewers, along with demographic and engagement data from multiple people simultaneously. Data is extracted and stored as a numerical log file with no images or video being stored.
How does it work?
Sensors located in display panels or inside mobile/tablet devices near product placements, scan the surrounding area. AVA is a computer vision application that processes video feeds in order to detect an arrangement of pixels that resembles a general pattern of a human face. AVA uses patterns such as pixel density around the eyes, nose, and mouth.
The facial features in the image are detected, and any other objects like trees, buildings, bodies, or animals are ignored.
Detection algorithms are based on a “learned” face pattern that has been trained on an audience database of thousands of face images. This allows the software to determine the gender and age of anonymous participants.
Each video frame is processed to detect the presence of human faces, and there is no database used to match faces to an identity, as would be the case with facial recognition. Non-identifiable information includes a person’s gender, approximate age, and facial expression.
What information is collected?
Anonymous information collected may include:
Total count of individuals
Demographic data such as gender and approximate age
Engagement data such as attention, duration time and number of glances
Viewer attributes such as the estimated distance and general position
Emotional expression (Facial Coding & Facial Imaging)
How is the information used?
AVA has no ability to recognize or identify anyone. The software gathers purely numerical data, none of which is personally identifiable information. No images are ever saved.
The anonymized data is aggregated by the software to report numerical statistics. The analytics generated by AVA software provides marketers and businesses valuable insights into what’s happening within the proximity of displays and other product locations in real time.
Understanding the dynamics of the viewing audience allows businesses to serve their guests better.
AVA also provides marketers the ability to assess the cause and effect of marketing, messaging, and map sales or other data against an audience.
The FTC Face Facts report recommends the following guidelines:
design services with consumer privacy in mind
develop reasonable security protections for the information you collect, and sound methods for determining when to keep information and when to dispose of it
consider the sensitivity of the information when developing products and services – for example, digital signs using facial recognition technologies should not be set up in places where children congregate.
The FTC staff report also recommends that companies take steps to make sure consumers are aware of facial detection technologies when they come in contact with them, and that they have a choice as to whether data about them is collected. So, for example, if a company is using digital signs to determine the demographic features of passersby, such as age or gender, they should provide clear notice to consumers that the technology is in use before users come into contact with the signs.
The Digital Signage Federation (DSF) privacy guidelines - which represents a wide range of companies from hardware and software vendors, to retailers and fast food restaurant operators - has recommended a set of privacy standards based on the internationally-used Fair Information Practices (or FIPs), which are incorporated in many privacy laws globally. The guidelines are voluntary recommendations.
Transparency: Companies should give consumers “meaningful notice” where the technology is in use;
Individual Participation: Consumers should have the right to opt out (with AVA, notice on site means consumers can choose to avoid the screens and sensors);
Purpose Specification: Published policies should explain how the collected data is used;
Data Minimization: Companies should limit their data collection and retention to only the minimum needed to achieve specified needs;
Use Limitation: Collected data should not be shared or sold for any uses that are incompatible the original purposes specified;
Data Quality and Integrity: If identifiable data is retained, consumers should have the right and mechanism to edit that information for accuracy;
Security: Any data collected should be secured;
Accountability: End-users should establish internal accountability mechanisms.
Kairos / IMRSV Use of AVA and Privacy
Kairos provides clear and unambiguous statements about the “anonymous” nature of the AVA processes.
No identifiable information is collected, retained, used, or shared using AVA.
Real-time video is scanned, analyzed and immediately discarded in the AVA process.
The aggregated anonymous data provides valuable, actionable insights for users.
Real-time processing means security and privacy risks are continually addressed.
Visibility and transparency. Vendors and the user community are encouraging consumer notice.
Respect for user privacy: keep it user-centric. Consumers should be empowered by this technology to participate and/or verify privacy claims.
Is face recognition the same as AVA?
No. Simply put, face detection detects human faces, it does not recognize who the person is. AVA has no ability to remember anyone once they have left the scene. Face recognition, as used in Kairos Face Recognition API, is a different type of imaging technology that searches for faces matched to images stored in a database. Face recognition can identify and remember a face even years after it was first recorded. AVA uses anonymized general traits and does not use face recognition. No face images are stored, and no identity information is matched. AVA uses some characteristics of the face to classify demographics (age and gender). Pictures of the person are immediately discarded.
It is important to understand the differences between facial detection and facial recognition in terms of the Kairos product line:
Kairos Face Recognition API incorporates both facial detection and facial recognition - identifying information is stored in the form of face-maps (the original pictures are discarded)
Kairos Crowd Analytics SDK uses AVA and incorporates facial detection - no identifying information collected
Kairos Emotion Analysis API API uses AVA and incorporates facial detection - no identifying information collected
IMRSV for Marketers uses AVA and incorporates facial detection - no identifying information collected