04 Jan 2017
Image processing technology initially developed by signal processing specialist CEVA.
Major imaging sensor manufacturer ON Semiconductor has agreed a licensing deal with California-headquartered CEVA to use the latter’s imaging and vision signal processing platform in automotive applications.
CEVA’s technology platform, developed with automotive advanced driver assistance (ADAS) products in mind, is said to provide new image processing capabilities incorporating embedded intelligence and machine learning.
Embedded imaging and machine learning were two of the key topics under discussion at the recent Vision tradeshow held in Stuttgart, Germany, where ON Semiconductor exhibited its wide range of CCD and CMOS sensor capability.
Ross Jatou, VP and general manager of ON Semi’s “automotive solutions” division, said in a release announcing the licensing deal: “The automotive industry requires cost- and power-efficient vision-based ADAS solutions to address the growing end-customer demand and safety regulations across all tiers of the industry.
“CEVA’s industry-leading vision processing intellectual property provides us with a comprehensive solution that enables us to integrate a host of innovative and intelligent system features into our ADAS product offerings.”
The two companies quote a recent report from the market research firm Strategy Analytics predicting that new camera applications and imaging concepts will see automotive camera demand to reach almost 200 million units in 2023, with a growing need for intelligent vision processors in related safety systems.
Rival analyst Yole Développement in France is even more optimistic, suggesting in November 2016 that by 2021 the average car will feature no fewer than three cameras - equivalent to "at least 371 million automotive imaging devices worldwide".
Wider field of view
In its report published last year, Strategy Analytics pointed to new detection requirements prompted by the European New Car Assessment Program (Euro-NCAP), including higher dynamic range, greater sensitivity for low-light conditions, higher resolution, a wider field of view and faster frame rates.
Those are deemed necessary to help protect so-called “vulnerable road users”, or VRUs, who Strategy Analytics say now account for nearly half of the deaths and injuries on Europe’s roads.
“VRUs crossing the road have presented auto makers with the challenge of widening the field of view in front windshield cameras – and with that, the resulting need to increase resolutions so that object recognition algorithms can still operate without compromising on operational range,” reckons Kevin Mak, a senior analyst at the market research firm.
He also points out that the increased camera performance will in turn demand increased image processing power, and that strategies are needed to ensure that the increase in performance requirements does not result in a drastic increase in costs.
New safety features
On top of that, new applications like long-range object recognition, traffic light detection, mirror replacement, and even detection of potholes and speed bumps, will all bring about changes in camera design.
“With advanced safety features set to be adopted across all price points of the automotive market, there is a strong demand for more cost-effective, flexible and scalable vision architectures,” stated CEVA and ON Semi.
“Efficient vision processing addresses computationally intensive imaging and machine learning use cases, including better low-light processing and the ability to run more powerful deep neural networks that can provide the accuracy and performance of tomorrow’s active safety systems.”
CEVA’s CEO Gideon Wertheizer said of the hook-up with the semiconductor and sensor chip manufacturer: “ON Semiconductor is a recognized leader in high-performance image sensors for the automotive market, and their selection of our vision platform is a strong endorsement of our vision IP for ADAS.”
“Their deep knowledge of automakers’ stringent requirements for performance, safety and reliability positions ON Semiconductor well to further extend their leadership in image sensors with the addition of our vision processing IP.”