Anonymous facial recognition is technically validated – We have a very high accuracy

A movie showing how we conducted the technical validation

A movie showing how we conducted the technical validation

The technical validation of Indivds unique technology for anonymized collection of data for all visitors in physical stores, which has been carried out in close collaboration with the researcher Karina T Liljedal, working at Stockholm School of Economics, is now completed with very satisfactory results.

– We achieved a very high accuracy when it comes to being able to re-identify the visitors. With continued tests, we have good hopes of reaching 95 percent accuracy. This is a very satisfactory result, says Dr Leonard Johard, CTO, one of the world’s leading AI researchers within Reinforcement Learning and architect behind Indivd’s new technology for obtaining anonymized data for the complete visitor journey, from entrance to conversion and revisit.

The technical validation of Indivd’s patent-pending technology for collecting anonymized visitor data and creating unique insights in retail is now completed. The technical validation was carried out in close collaboration with the researcher Karina T Liljedal, working at Stockholm School of Economics, during two days in May 2019 with the help of 230 volunteers, a grocery store and a coffee shop. The study achieved an accuracy of about 85 percent re-identifying the returned visitors without storing personal data.

The validation was planned in collaboration with researcher Karina T Liljedal and is a part of the scientific research project, which Karina T Liljedal is currently carrying out.

– We used two different cameras during the study, one older model and a completely new camera. This means that we can safely state that the technology works well with both the older and newer cameras installed in the stores today, says Dr Leonard Johard, CTO, world-leading AI researcher and architect behind Indivd’s new technology for obtaining anonymized visitor data.

95 percent accuracy with anonymized data

Indivd will test and refine the technology together with our first store early autumn this year. A  pilot project based in Stockholm. The purpose of the pilot is to collect anonymized data in while changing the environment and supply in order to study, in more detail, the effects.

– We have previously known that our accuracy of unanonymized re-identification, ie re-identification with the help of regular facial recognition, is 99.75 per cent, and now we know that using our unique technology for anonymized facial recognition, we will be able to achieve an accuracy of over 85 percent. We also know that the more data colleted will increase our accuracy since all noise will be automatically sorted out. Based on that fact we have very good hopes that we’ll quickly achieve a 95% accuracy for re-identification without storing personal data. This is a very satisfactory result, says Dr Leonard Johard.

The importance of being able to re-identify

The revisit frequency is considered to be something of a holy grail when it comes to measuring visitor insights.

– The rest of the valuable KPI’s in a store is easy If you can understand the revisitation frequency. We need to be able to re-identify the visitors in order to understand behaviors since behavioral insights are aggregated data from visitors over time. If we can measure return visits without processing biometric data or storing personal data, then we can also understand changes over time linked to the strategies that the stores use. If the visitors revisit or not makes us able to understand how the changes affect the visitors, says Dr Leonard Johard.

A huge potential without processing biometric data or storing personal data

Facial recognition is considered to be the best technology to understand what actually happens in the stores. Regular facial recognition, however, has major problems to cope with the tough and necessary legal requirements regarding privacy.

Indivd, however, through its new, patent-pending anonymous technology, use cameras to collect demographic data, local data, behavior data, and interest data without stepping over the boundaries. Our technology doesn’t process biometric data or store any personal data. The collected anonymized data is impossible to trace back to physically living people.

– Our technology can re-identify all visitors over time and place. Indivd is compatible with new cameras, as well as with the 245 million cameras that are already used in stores today and in other environments around the world. There is a lot of potential here, ”says Dr Leonard Johard.

How we conducted the validation

Exactly 230 randomly volunteers participated in the study. They signed an agreement where they gave their consent for us to collect their personal data and where asked to follow a specific route, first a restaurant and then a store. All participants were offered a coupon as compensation with free crepes and coffee at the restaurant in question.

The participants also received a bonus coupon, upon acceptance,  with a specific bar code. A bar code which linked the participant to the data from the manual study with data from our anonymous facial recognition. 

The participants were registered both manually and with the help of cameras when they first entered the restaurant and when they were subsequently re-identified in the nearby store. This made it possible to synchronize data from the manual study and with our new technology. 

The study was planned to reflect a real scenario with the use of old and new camera models, different lightings and various angels.

For more information contact:

Leonard Johard, Head of AI & Lead Scientist,

Karina T. Liljedal, PhD Center for Consumer Marketing, Center for Retailing, 0046 734 612 484,

[et_bloom_inline optin_id=”optin_3″]

Share This

Share this post with your friends!