The Eye of AI #8

Computer vision is a truly transformational technology for the retail industry, but it has applications that touch nearly every aspect of our lives. This is our weekly summary of what happens when cameras, computers and great ideas get together.

Fakes Spotting

As counterfeits flood the luxury goods market, brands and law enforcement agencies are turning to computer vision to help identify fakes. With the value of genuine products projected to hit $385 billion by 2025, technology is helping brands authenticate products with precision, saving them billions. Counterfeit goods are widely used by criminal gangs to launder money, but correctly identifying seized fakes is time-consuming, specialist work. Using computer vision helps prosecutors quickly assemble tighter cases and can even help connect activity by identifying similar batches of fakes. In the burgeoning secondary market, re-sale companies are integrating computer vision to enhance their verification processes. Across the luxury spectrum, from sneakers to watches, the technology is becoming essential in keeping counterfeits at bay and ensuring trust in high-value transactions.

All's Weld that Ends Well

Using incredible, localised heat to temporarily turn metal into liquid is hard, noisy and dangerous work. Car manufacturers have long used robots to improve the production quality of vehicles, but with a global welder shortage, a new version of welding robot is using computer vision to learn from previous generations, delivering consistent, reliable results and setting the stage for more AI-driven automation in the future. Autonomous welding is evolving fast, with Novarc Technologies leading the charge. The company’s new system uses real-time vision processing and AI to continually improve weld quality through data collection and model enhancement. By automating the complex pipe welding process, the robot ensures high-precision, uninterrupted welding while adapting to inconsistencies. This technology meets stringent industry standards across sectors like aerospace and shipbuilding and reduces worker exposure to welding hazards.

A Mission to Mars

The greatest journeys start with the first step, or in this case the roll of a wheel. In a quarry just outside Leighton Buzzard in England, the four-wheeled rover "Codi" is undergoing trials that could ultimately see it being sent to another planet. Showcasing its ability to autonomously collect sample tubes using a robotic arm and advanced computer vision. Codi navigates with 10 cm precision, constantly mapping the terrain and using stereo cameras to plan its route. Once parked, its mast camera detects and retrieves sample tubes, mimicking NASA's Perseverance rover's Martian soil collection. The rugged quarry, with its rocky slopes, provides a dynamic testing ground, helping to advance rover capabilities in preparation for future space missions.

Logging Logging

Traditional timber measurement, a slow and labor-intensive process, faces significant limitations in accuracy and safety, particularly in harsh conditions. The need for faster machine learning execution and better processing power led to a transition from older platforms to a modern AI-powered system. This new system, powered by advanced AI technology, significantly improved efficiency, detecting up to 1000 log ends per frame in just 200ms. The shift also enhanced accuracy, reduced false positives, and improved safety by automating measurements, allowing workers to stay in vehicles. The compact, low-power solution now supports real-time, precise timber measurement, with scalability for future AI advancements.

Build Your Own Vision Transformer

If you're feeling the need to get amongst the nuts and bolts of computer vision systems, here's a handy guide to building your own Vision Transformer (ViT). The ViT underpins a whole host of computer vision systems by taking the idea of a transformer, the T in ChatGPT, and training them on visual data. Transformers were initially developed to improve machine-based language translation and introduced the idea of Attention into machine learning - essentially, giving computers the ability to determine what is important. In language translation this may mean giving particular importance to a particular group of words; in ChatGPT's case it may be paying attention to how a particular word changes the context of a question. The important thing to know is that transformers work because they work, but nobody really understands why.

That's everything for this week. Please keep an eye on the SAI Group blog for everything that we're thinking and talking about, including a new article this week about the things to consider when deploying computer vision in retail stores.

Got some cool tech to share? Whether its your own project, or just something you saw and thought "I want people to know about this!", let us know about it and we'll include it in upcoming editions.