Overview / Usage

Analytics: It is not uncommon for retail spaces to capture customer footfall data. The methodology to gather however relys on methods like manual counters at entry gates, POS systems or recently BLE beacons. These systems are not able to reflect the demographics of the foot-fall. Retail owners would like to know for instance_ How many women between the age 30-50 visited the apparels section?_

Flux Cloud is capable of tapping into existing CCTV infrastructure and capturing foot-fall metrics linked with age, gender, focus and even allow customers to define attributes to be identified using custom models.

Targetted Advertising: Retail spaces fail to deliver personalized experiences. Especially personalized recommendations at par with e-commerce services. Flux allows retails spaces to digitize the advertising experience and personalize ads based on the customers within viewing span of the digital billboard. Example: Display 20% discount coupon codes on prescription glasses for customers wearing glasses.

Flux Boards allow retail spaces to install digital billboards coupled with thin-clients capable of taking over the edge decisions on which ads to show considering the customers within the board's viewing span.

Methodology / Approach

Flux aims to provide two complimentary capabilities to customer. Analytics and Targetted Advertising. The technical implementation of both these features has a significant overlap - using computer vision to extract personalization attributes like age, gender, emotion, etc. The contrast being that Targetted Advertising on Flux Boards relies on realtime inference, Analytics on Flux Cloud is capable of batch processing CCTV footage relayed over by on-premise gateways to gather similar metrics.

It is architecturally and operationally advantageous to be able to reuse implementations across two polarized infrastructures. A thin-client and a cloud service. Intel OneAPI provides a standard interface for building solutions that can leverage a diverse set of accelerators.

Flux uses OpenCV and trained models to identify human features such age, gender, facial attributes through a mounted camera. We are working to extend the models to identify apparel and accessories. Flux uses the captured features to filter through product listings and presents products that result in optimum amount of engagement.

The solution is roughly based upon Convolutional Neural Networks for Fashion Classification and Object Detection http://cs231n.stanford.edu/reports/2015/pdfs/BLAO\_KJAG\_CS231N\_FinalPaperFashionClassification.pdf

Technologies Used

Intel® AI Analytics Toolkit, Azure (Intel HPC VMs), OpenCV, Tensorflow, Electron

Collaborators