Edge AI and Vision Insights: January 29, 2025

LETTER FROM THE EDITOR

Dear Colleague,

Don’t miss your chance to save 25% on registration for the 2025 Embedded Vision Summit, coming up May 20-22 in Santa Clara, California! Register with code SUMMIT25-NL to take advantage of this price.

We’re also excited to announce our first batch of expert speakers:

Ravi Kota, CEO of AI Tensors
Lazar Trifunovic, Solutions Architect at Camio
Francesca Palermo, Research Principal Investigator for EssilorLuxottica, and
Matt Merrifield, CTO of The Nature Conservancy

These speakers will soon be joined by 100+ others, all focused on building products using computer vision and AI, so stay tuned!

There are a variety of pass options available to fit your budget—made even better with this discount! Secure your spot at the best price today. We look forward to seeing you and your team at the Summit.

Brian Dipert
Editor-In-Chief, Edge AI and Vision Alliance

INFRASTRUCTURE INSPECTION

Edge AI Optimization on Rails—Literally

In this talk, Matthew Pietrzykowski, Principal Data Scientist at Wabtec, shares highlights from his company’s adventures developing computer vision solutions for the rail transportation industry. He begins with an introduction to the types of machine perception problems encountered in this unique industry. He then shares insights gained during his company’s implementation of a perception system for rail auditing. In particular, Pietrzykowski discusses the challenges faced in implementing multiple optimized CNNs in a constrained compute system, and how his company addressed these challenges. He also explains trade-offs associated with using classical homographic techniques in combination with neural networks.

Innovative Applications of Computer Vision for Power Utility Infrastructure Inspection

In this presentation, Vikhyat Chaudhry, Co-Founder, Chief Technology Officer and Chief Operating Officer of Buzz Solutions, delves into an innovative application of computer vision for power utility infrastructure inspection. He begins with an overview of Buzz Solutions’ technology, emphasizing the use of computer vision for addressing critical challenges faced by power utility clients. He highlights key challenges his company has faced in using computer vision in this sector, including a lack of publicly available data and a lack of standards for data capture across utility clients. Chaudhry then explains how his team overcame these challenges using innovative techniques such as generative AI. He shows the results of these techniques by showcasing Buzz Solutions’ PowerGUARD solution, a vision edge AI system for substation monitoring. Finally, he shares some of the key learnings and insights his company has gained from applying computer vision in the power industry.

IMAGE SENSOR AND CAMERA TUTORIALS

Seeing Through Machines: A Guide to Image Sensors for Edge AI Applications

Imagine a robot navigating a busy factory floor, a self-driving car detecting obstacles on the road or a security system accurately recognizing faces. These advancements in artificial intelligence are powered byCMOS image sensors. With a vast array of sensor options available, choosing the right one for your specific application can be challenging. In this talk, Armita Abadian, Advisor at SEEdar Consulting, demystifies the world of image sensors, exploring how they work and uncovering the key performance criteria you need to consider for projects in automation, robotics and beyond. You’ll be able to navigate the sensor landscape, equipped with the knowledge to make informed choices and unlock the potential of edge AI.

Introduction to Cameras for Embedded Applications

In this presentation, Brian Rodricks, CTO of SensorSpace, introduces the essential features of cameras for embedded vision applications. He explains lens mounts and camera interface options such as MIPI, USB and GigE. He explores sensor features and trade-offs, including rolling shutter vs. global shutter, monochrome vs. color, RGB vs. RGBW and pixel size vs. array size. Rodricks also examines trade-offs between field of view and distortion, as well as performance trade-offs such as dynamic range vs. sensitivity, pixel size vs. resolution and frame rates. Finally, he addresses the crucial aspects of price vs. performance and control, covering access to register settings, raw data and software.