Video processing platform reduces costs, makes systems fail-safe

Video processing platform reduces costs, makes systems fail-safe
Technology News |
The fact that cars can "see" is nothing new. A new technology platform from computer vision and AI expert Videantis GmbH now brings more intelligence to image processing. At the same time, the video capture process becomes fail-safe so that it can also be used in safety-relevant applications - without conventional lockstep mechanisms.
By Christoph Hammerschmidt


For three years and with a budget of €50 million, the European R&D project PRYSTINE (Programmable Systems for Intelligence in Automobiles) involved about 60 project partners working together to develop a Fail-operational Urban Surround perceptION (FUSION) based on radar and lidar sensor fusion and control functions to eventually enable safe automated driving in urban and rural environments.

The results of this project contribute to meeting advanced functional safety requirements for embedded Videantis-based multiprocessor systems up to ASIL D according to ISO standard 26262. Compared to conventional lockstep architectures, more than 50% of costs can be saved by reducing the silicon area.

The cost reduction is achieved through run-time fault detection schemes consisting of core self-test modules and a result monitoring software layer (RMSL) applied to the fine-grained and highly scalable Videantis multiprocessor system. In this way, faults can be detected at runtime and processing can continue, excluding the faulty resources without the need for duplicate hardware. For example, a multiprocessor system with 32 Videantis cores can be transformed into a fail-safe processing platform with a silicon area overhead of only 3%.

“PRYSTINE covers the most important aspects of autonomous driving: performance, efficiency and above all safety. We are proud to have contributed with our versatile processing platform to implement fail-operational functionality in the most cost-effective way,” says Dr. Hans-Joachim Stolberg, CEO/CTO of Videantis.

The Videantis v-MP6000UDX processing platform is specifically designed for use in fail-operational applications such as highly automated driving. With its scalability, it can cover the entire range from intelligent image, radar and lidar sensors (1 to 16 cores) to high-performance AI inference computers (>100 cores). The uniform architecture enables the implementation of various functions: Video coding, image or graphics processing, computer vision, deep learning using a variety of network topologies or even control functions. The use of redundancy, self-testing and other control mechanisms enables customers to build secure systems according to ISO26262 up to ASIL D, requiring less silicon space or hardware overhead than traditional lockstep architectures.

Related articles:

AI platform for autonomous cars reconfigures on the fly

Drones learn autonomic navigation from cars and bicycles

Fail-operational architecture for automated driving covers L3, L4

Linked Articles
eeNews Automotive