Summary

Solving software and cybersecurity challenges for robust smart camera applications on Xilinx Kria SoMs

Vision systems are rapidly evolving using algorithms based on Machine Learning (ML) to enable detection and complex elaboration in often uncontrolled environments. Total system performance is determined by carefully matching the ML inference subsystem with the Pre- and Post-video processing, ideally integrating in a single device to reduce power consumption and total cost of ownership. Addressing such a system from a software perspective is usually the biggest challenge.

In addition, connectivity on these camera systems coupled with their often accessible physical locations imply exposing devices to world-wide security threats. The attack surface is mainly software, but the integrity of the software can only be ensured by a root of trust in the hardware along with remote attestation of the software and firmware. Similarly, the protection of cryptographic material requires hardware security features. Thus, managing security for Industrial IoT involves setting up security requirements very early in development, achieving in the process of a proper secure by design approach.

During the webinar we present a tutorial based on Xilinx Kria SoM that explains how to acquire video from an camera module, preprocess it using the programmable logic, run inference using pretrained and to-be-trained models and stream the results. We will do all of this using an integrated GStreamer pipeline without HDL coding, and with all the security requirements and aspects taken into consideration. Kria SOMs ability to support secure and measured boot with a hardware root of trust to build a IEC 62443 compliant system is also covered.

Who should attend?

• Software Engineers and Product Managers who are exploring the norms to comply with or the requirements to cover their vision AI technology roadmap
• Cybersecurity Managers or IT Managers who desire to cover the “product” part within their approach.
• R&D Managers who are actively developing ML for Vision Applications

Agenda

What you will learn:

• Added-Value of ML (Machine Learning) for your ecosystem with use case examples
• What is the risk if you don't consider implementing security into your Industrial IoT system
• Key elements to avoid mistakes and best practices to develop Secured Software & ML on Kria SoMs

Speakers

Chetan Khona
Director Industrial, Vision, Healthcare & Sciences
AMD

Chetan Khona is the Director of Industrial, Vision, Healthcare & Sciences markets at AMD, based in New York. Before joining AMD in February 2022, Chetan was captivated by the creative potential of Xilinx devices to solve embedded design challenges in a fundamentally differentiated and efficient manner while a student at the Georgia Institute of Technology.

Chetan has continuously been involved with Xilinx now AMD as either a customer, engineer, salesperson, or marketer for over 25 years. These days, Chetan’s efforts have been focused on driving AMD roadmap and strategies to address the rise of AI and other embedded processing technologies at the IIoT edge supporting intelligent and adaptive factories, hospitals, and infrastructure applications. Chetan holds a Computer Engineering degree from Georgia Tech and a MBA from the University of Rochester.

Julien Bernet
Security Manager
Witekio

Julien is a vulnerability research engineer.

After a PhD in theoretical computer science, he’s been working for various security labs with a focus on software security for embedded devices and smart cards.

He’s been involved in many security evaluations and has a soft spot for the use of formal methods in software security.

He is now Security Manager at Witekio, in charge of the development of the internal security framework and defining the security architecture for developments.

Mario Bergeron
Machine Learning Specialist
Avnet