From Image to Vision: When Software Disrupts Hardware

Estimated reading time: 3 minutes

Vision systems are becoming increasingly important and under this context, Yole Développement (Yole), the “More than Moore” market research and strategy consulting company, investigated the software world to analyze the tight links between embedded software and hardware at the technology and market levels. 

The first segment image processing, very close to the hardware, inscribed inside standalone FPGA or ASIC chips, or integrated into more complicated architectures.

“This layer, not often considered, is the most important step in any image treatment after image acquisition by pixels,” comments Yohann Tschudi, Software & Market Analyst at Yole Développement. And he adds: “Image processing, realized in the ISP , has a quite simple function: the transformation of the sensor’s signal to an understandable image for the human eye."

The second layer is completely different, with much more diverse and complicated functions. Under its new report, Yole focused on embedded software and, more precisely, inference software derived from the latest AI methods. This kind of technology necessitates a lot of memory and computing power.

What are the software technologies? How do they impact the hardware? Which hardware is impacted? What kinds of markets are affected? And how will they evolve? Yole’s analysts offer you to discover the embedded software industry and its impact on the hardware side in vision systems at the technology and market levels.

AI has completely disrupted hardware in vision systems, and has had an impact on entire segments, like Mobileye has in automotive, for example. Image analysis adds a lot of value and image sensor builders are therefore increasingly interested in integrating a software layer to their system in order to capture it.

Today, image sensors must go beyond taking images – they must be able to analyze them. However, to run these types of software, high power computing and memory are necessary, which led to the creation and development of vision processors. ISP market offers a steady CAGR of 6.3%, making the total market worth US$4,400 million in 2017. Meanwhile, the vision processor market is exploding, with a 30.7% CAGR and a market worth US$653 million in 2017! Today, optimization requires software and hardware to be developed in parallel. Depending on the issues and specifications, companies can invest more in hardware than software or vice versa. However, software is easier to specify, tune and update, and so its growth is more important than hardware. The AI market is therefore expected to reach US$35 billion in 2025 with an estimated CAGR at 50% per year from 2017-2025.

“At Yole we carefully evaluates ISP and vision processor market shares and their evolution in order to correctly understand how AI technology impacts the hardware”, explains Yohann Tschudi from Yole. This market has been divided in two different business models: IP companies, which don’t have physical products, and hardware companies, which sell the processors physically. The leaders are pretty easy to identify for each category. ARM and Synopsys lead the IP segment and Omnivision, Mobileye and On Semiconductor lead the hardware segment.

The AI market, particularly in vision systems, is new and still moving, with hundreds of startups created each year. It has no clear leaders but a lot of highly specialized companies.

AI technologies promise a bright future in many areas, with rapid software and hardware progress. In autonomous vehicles, AI allows cars to understand the world around them, predict trajectories, communicate and drive. This has led to the development of sensor fusion boards. For example, NVidia’s Drive PX boards provide very high performance computing and memory, giving the ability to compile information from many completely different sensors. In surveillance and security, face and iris recognition have never been as powerful, entering the consumer world through the iPhone X this year, and behavior recognition is on track.