At CES 2020, Ambarella’s latest demo version was shown, among other things, which presented various solutions with CV2 and CV22 and new platforms based on the newly announced SoCs for vehicle cameras CV2FS and CV22FS.
For readers unfamiliar with Ambarella, the company has become known for its success in providing the silicon in portable solid-state camcorders as well as in sports cameras such as the GoPro Hero range. Over the years, the company has shifted its product towards more specialized use cases and now claims to be a leading provider of IP video security solutions and is also a leader in providing solutions for automotive platforms.
Continued success with CV2 and CV22
The CV2 and CV22 solutions were presented at the CES last year and continue to be important solutions and offers for the company for 2020. Among the more interesting demos that they presented this year was a direct comparison with a competitor solution that Performance and energy efficiency benefits of the CV2 platform:
The CV2 development board was used against an Nvidia AGX running an object recognition workload. Both platforms showed similar performance at 60 fps (~ 13.2 ms for the AGX versus 16.9 ms for the CV2 in terms of inference time), although the Nvidia platform consumed 32 W of power compared to just 6.9 W for the CV2 demo. We looked at the AGX last year and found that similar inference workloads with the Nvidia demonstration software are expected to consume 13-16W of power. The AGX may not have been used optimally in the Ambarella demo implementation.
CV22 Dev Board
The company presented further partnerships with various companies using the CV2 and CV22 platforms, including working with AWS SageMaker Neo platform to train ML models in the cloud and using the Ambarella CV SoCs Edge devices.
CV22 Dev Board
In addition to a partnership with AnyVision for the provision of retail analyzes (heat maps, traffic analysis, person recognition, recognition and counting in retail and retail stores), a very interesting demo with Mercedes-Benz as part of a so-called “Cargo Recognition and Organization System” (CoROS), where there is a CV2 device and a camera at the top of the back door of a van that can be used to scan loaded and unloaded parcels. When a particular package is scanned during loading, it can use LED strips on the shelves to mark the most optimal shelf space in the delivery van and optimize the arrangement of the packages in order to achieve an optimal loading and unloading experience depending on the shipping route. The implementation of the system was extremely straightforward and required only a single camera with a higher resolution in order to be able to effectively read out package barcodes. It definitely felt like a killer use case for computer vision solutions.
Automotive: further development and new ASIL B CV2FS and CV22FS SoCs
With regard to the automotive display cases, we have seen further improvements on the software side of automotive products in cooperation with companies such as HELLA-Aglaia.
(embed) https://www.youtube.com/watch?v=zwAyl5RQ3PQ (/ embed)
Front-Facing ADAS System Demo (newer 360 ° system was in another car)
Although we couldn’t test it during our daytime CES schedule, the most interesting demonstration the company had shown was a fully autonomous vehicle demo, using only CV2 chipsets and various camera systems. Ambarella was proud that the system worked both day and night. The latter is much more complex to implement in a pure CV system without LIDAR.
The new product announcements this year were in the form of the CV2FS and CV22FS – essentially, they are brand new designs based on the CV2 and CV22 functions and now fully meet the ASIL B requirements for functional safety and qualification for vehicles like AEC-Q100 offer grade 2 compliance (-40 to + 125 ° C operating temperature).
CV25 For standard cameras and consumer devices
No recent announcements were displayed on the consumer camera page. Therefore, the CV25 remains Ambarella’s main product for consumer and surveillance camera applications.
Merged RGB-IR image captures an illuminated subject (left) compared to a simple non-illuminated subject (right)
Another interesting development was the announcement of a partnership with ON Semiconductor to bring a new RGB-IR camera sensor to the market that can capture information in both the regular RGB color spectrum and the IR spectrum. Ambarella’s ISP is able to support the format and merge the data, creating some very interesting new features related to low light detection. The function appears to be another killer use case that will be implemented in security cameras in the future.