Magna says AI-boosted thermal sensing is more effective, cuts back on false automatic emergency braking activations

Magna International Inc. is pitching automakers on a new thermal sensing product that processes information with the assistance of artificial intelligence.

The supplier says its system reduces the number of unexpected automatic emergency braking activations.

Magna's new AI-based convolutional neural network with thermal sensing was named in August as a finalist for the Automotive News PACE Pilot awards. The awards recognize precommercial and postpilot innovations in the automotive or future mobility spaces.

With the help of AI, Magna trained its new thermal sensing algorithms on more than 70 million frames of images taken by sensors on roadways, said Richard Seoane, Magna vice president of operations and business development in thermal products.

Pedestrians, animals, cyclists and other vehicles are marked on the images, which Magna's algorithm is trained to recognize more accurately than today's standard systems.

"By adding this thermal camera to that sensor suite, it gives the system more information to make intelligent decisions without going into an automatic emergency braking situation," Seoane said.

Unwanted automatic emergency braking activations are a source of unease for automakers as they work to meet a new 2029 NHTSA mandate requiring the systems be in new vehicles. Mistaken activations can create front-to-back crashes by trailing cars. Drivers don't anticipate the sudden stop.

"Automatic emergency braking is not a comfortable situation," Seoane said. "It's a hard brake, and depending on the situation, you also have fear that you might hit a pedestrian."

The system helps cut down on errors that could lead to unwanted automatic emergency braking activations, Seoane said. A standard algorithm for thermal systems used in today's vehicles might mistake a fire hydrant giving off heat at night after a hot, sunny day as a small child, he said.

"With an object-based algorithm, it'll give you a false positive saying that's a pedestrian," Seoane said. "With the convolutional neural network, it doesn't recognize that as a person."

Likewise, Magna's new system would be able to detect if a person bending down to pick up something they dropped is in fact a pedestrian. Older systems might not make the same conclusion if they cannot easily recognize the person's head or shoulders.

"If you have someone laying on the ground, it wouldn't classify it," Seoane said. "So there were a lot of missed detections."

The new algorithm is trained to recognize people in many different scenarios, including when they are looking down at their phones or carrying objects, Seoane said.

"The beauty about training that data is that you can train it for different types of poses," he said.

This is the latest thermal system developed by Magna, which has supplied more than 1 million thermal systems to automakers since 2003. It is meeting with customers to gauge interest in the new product and to discuss how it can be implemented into existing safety systems, Seoane said.

Magna has been making thermal systems since 2003, but this is the latest iteration. The older systems didn't utilize an AI-based convolutional neural network.
The company hopes the new capabilities will help thermal sensing to become part of the more robust advanced driving assist systems of the future.

"A thermal sensor can only do so much independently," Seoane said. "But when you start to pull different sensors together and use AI-based algorithms, you can really develop a strong ADAS system."

Magna ranks No. 3 on the Automotive News list of the world's largest suppliers, with global sales to automakers of $42.8 billion in 2023.