Musings on human vs. computer vision

When I founded Sentispec, it was from the basic notion that in the animal kingdom, the one sense that really sets humans apart is our eyes. Not because our eyes are mechanically superior, but because of the immense processing power positioned immediately behind the eyes — our human brain.

Human brains are incredibly powerful — even our fastest super computers are still 1000 times less powerful (and are not at all mobile).

As such this begs the question — what are the manual processes which one might improve if just one could observe them 24/7? Whereas this would be unviable using human labour, it might just be possible using computer vision. Simply place a camera to observe the manual process and give you back the statistical results.

And thus Sentispec was born.

In the field of AI, the area which has always held the most potential data is computer vision, and it is only now, with good and cost effective cameras and processing power, that we can truly start tapping into the power of it, by placing cameras in the right places and realtime processing the results into business insights.

Take for example our Automated Stock Taking solution — we generate some 200–300 MB of data per second, which would have been unthinkable to process in real time at an affordable cost just a few years ago.

Stock taking is a great example of a tedious manual process which can be optimised with computer vision in a quick affordable way, and with a Return on Investment within 3 months.

If you run a pallet hotel of, say 100,000 pallets, you will likely count these 4 times per year at a manual productivity of 50–100 pallets/hour. With a labour cost of €25/hour this cost is around €100–200,000 annually.

With the Sentispec AST solution we max out north of 3000 pallets/hour. Yes. 3000. The short version of this is an 80–95% cost saving, or even better, the ability to free up several full time people from cost-only work and reallocate them towards revenue generation. Simultaneously, you get a higher level of stock transparency and spend less time looking for lost pallets. And "all" we did was put a high speed camera on a pallet and a forklift, and process the data with AI.

Beyond Stock Taking, we have solutions available for ensuring Loading quality, and optimising Fill Rates across an entire network. We are looking at solutions for automatically scanning goods out at gate (saving 2–3 minutes per trailer load), and generating performance and productivity metrics for Picking operations.

That's probably enough rambling for one sitting so let me end on the immortal words of Captain Jean-Luc Picard:

"To boldly go where no one has gone before!"