There's an interesting contradiction in society today. On one hand, brick & mortar retail is dying. Blockbuster, Sears, Zellers have each said their farewells. On the other hand, Internet commerce giants Amazon and Alibaba have recently invested millions into opening brick & mortar stores of their own.

Why are Amazon and Alibaba entering the same space their core services sought to destroy?

They see something irreplaceable in brick & mortar retail: tangible, visible data points. You and I, we are walking goldmines of data, and all of our value is wasted in the hands of Walmart and Costco.

But it doesn't have to be this way. With MannaGer, we've built a smart, data-driven system to manage brick & mortar 2.0.

Using CCTV footage in real time, we extract demographic insights, reveal trend comparisons, automatically deploy workers and allocate resources.

In the long term, we present concrete evaluations of marketing campaigns, predictions of consumer purchases, and recommendations on hiring and inventory.

We may just look like a very cool data company right now. In reality, we are trendsetters. By employing AI to automate management rather than individual tasks, we will be the ones to push every industry forward, starting with MannaGer for retail.

We invite you to join us in this journey.

Our awesome technologies

After figuring out the most interesting features we can add, we realized that the biggest part is tracking people in the video and analyzing them. We started with using Google Cloud video intelligence API.

On top of that, we custom-built a system that can determine whether two detected people in two different frames are the same person. Then, after finding people, we pass their faces to the Face Plus Plus API which gives us demographic information: ethnicity, gender and age. In the end we created our script that gets frames of video and feeds them to the system to make it work.

There are many challenges in trying to make this work real-time and we tried to make it as close as we can. It is hard to make the algorithm robust in detecting people. Testing was another issue, we made our own model of a shop at Stanford earlier today and took some videos for testing the different features of the algorithm.

We started last night from ideation. Today, we have a behavior detection algorithm, with gender, age, ethnicity and emotion, which can analyze people's behavior in shops with 10 fps speed.

Share this project: