101 Jefferson Dr, 1st Floor
Menlo Park CA 94025

(408) 623 9165

info at jpdata dot co
sales at jpdata dot co

Computer Vision Has Come A Long Way

I started on my analyst journey with a report about Computer Vision back in 2015. This was the time when classic techniques ruled, and the best AI could do was to recognize a cat versus a dog. IoT was hot, and semiconductors, in general, had lost their charm as the darling of the VC community. Embedded Vision Summit was the only conference focused on the commercial aspect of computer vision and hardware.

Fast forward five years, and the word ‘Embedded’ has become ‘Edge’ (although no one agrees on the correct definition of edge, but we’ll leave that aside for now). Classic vision techniques have become obsolete (except for pre-processing), and neural networks have taken over tasks such as object detection and classification. Everyone is talking about how wonderful (and harmful) AI is. Computer vision has gone in household devices such as smartphones and enabled sci-fi-like applications.

I was at the Embedded Vision Summit conference recently and it was clear that computer vision has come a long way since then. The conference itself has increased its attendance by over 8x, from 160 attendees in 2012 to 1000+ in 2023, keeping up with the growth in computer vision professionals. The conference was full of exciting technical and business presentations and a crowd-pulling exhibitor section. It was refreshing to see every company with a demo as opposed to slides and freebies. Some companies were on their second-generation products. There was a mix of companies from software, hardware, cloud, and services companies suggesting that the ecosystem is developing.

Looking back, I can take pride in stating that many of my predictions have come true. Of course, I am not claiming that I knew the winning lottery numbers, but the numbers are holding up. The prediction for CV software and hardware market was expected to reach $33 billion in 2019 in that report. (There were several assumptions made in this calculation, and happy to chat if someone wants to understand how I arrived at the number). Automotive, consumer, and industrial markets were expected to be the three largest contributors to computer vision. T

Indeed, computer vision has penetrated all those three markets and has done very well. Cameras have become an integral part of automotives, with ADAS, in-cockpit monitoring, and even after-market parts. Blind spot detection and lane monitoring has become a common feature in many high-end automotives. While we still haven’t reached the level 4 and 5 automation that will increase the market drastically, we are on our way.

In the industrial market, robots are increasing their capabilities but still haven’t reached the potential they promised. Drones are increasingly using vision for navigation. Industrial QA systems have risen in popularity, and cobots are starting to make inroads. But overall volume has still not hit the millions mark. It will be another five years before we’ll see that volume reaching that level in the industrial market.

The consumer sector, driven by smartphones is increasingly using cameras and vision. Google’s technology that erases unwanted objects from a picture has become a key advertised feature. There’s very little penetration in other consumer devices such as TV or set-top boxes though, but devices such as smartphone, PC, tablets add up to a very large market for computer vision.

Video analytics for safety and security as emerged as a popular application. The retail industry is increasingly using video feeds for business and marketing analytics.

Overall, computer vision has done well in edge market – consumer and automotive have been key drivers and industrial to some extent. Many other sectors have emerged, such as agriculture, retail, and toys, to name a few. In fact, vision is on its way in every device in one form or the other.

What about computer vision within enterprises and data centers? This was not even a consideration back in 2015. It was questionable if the latency and performance of running the application on a server would match up to the requirements. There’s certainly been progress – non-real-time applications, such as video-based business analytics, have been steadily progressing. Applications such as image tagging and image search have done well. OCR usage is on the rise within enterprises.

I could go on and on, but you get the point. If you search for ‘computer vision’ to get a count of companies in Crunchbase, you get more than 2 million hits! This is not to say that there are 2 million computer vision companies, but it tells me that all those companies use the ‘computer vision’ word somewhere in their description.

Sometimes, it almost feels like computer vision is a mature field, given the hoopla surrounding generative AI. The world is moving rapidly, and so is vision. I thoroughly enjoyed meeting industry colleagues and familiar faces at the Embedded Vision Summit. We’ve come a long way in the past five years, and I hope that we continue to move forward to improve human life, not destroy it!