Filled under: AI/ML, Architectural, Commercial, Computer Vision, Data Visualization, Installation, IOT, Python

Video Analytics for Retail

2017-18, I led a cross-functional R&D team at our company that partnered w/ R&D at Xylinx, Intel, Google, and XNOR for next generation retail smart camera design.

I first demonstrated initial ‘back end’ AI/ML models interactively at NRF annual conference in NYC in 2019.

I then deployed an Intel FPGA based solution as pilot in San Jose March 2019 (at the time, FPGA was more performant per watt than a GPU)

goal was to measure throughput of 4 sales channels (online order w/ pickup, drive-thru) using new and existing security cameras.

Pilot Summary:

object classification and tracking in a commercial setting: Vehicles and people via 14 streaming cameras w/ model ensemble (object detection, tracking)
Then re-id across cameras was added, and then some demographics and other KPI for interest to retail for final AI/ML pipeline.

Also see an Intel Blog on some of the outcomes of the R&D efforts

Tools : OpenVino , C++, Python, FPGA

following images roughly in order from planning to prototype to pilot.