AutoSens: Nodar’s Plan to Make Park Assist Tech Better, Cheaper, Cleaner

CEO Leaf Jiang told an AutoSens USA 2024 audience how his company’s untethered stereo camera software could be used in place of today’s ultrasonic sensors.

July 17, 2024

Sebastian Blanco

Using one of five stereo pairs – here, the two rear-facing cameras – Nodar claims sub-millimeter accuracy within 4 meters (13 ft). (Nodar)

Park assist technology seems like low-hanging fruit in the assisted and automated driving space, but anyone who’s attempted to use one of these systems might have quickly realized that current solutions aren’t as easy to use nor as functional as they could be.

Nodar CEO Leaf Jiang speaks at AutoSens USA 2024 about using untethered cameras as stereo pairs to make park assist systems cheaper and better. (Sebastian Blanco)

Speaking at AutoSens USA 2024, Nodar CEO and founder Leaf Jiang said drivers have been complaining about self-parking features for years, but it remains a $2 billion market that is expected to grow at 17.5% CAGR between 2023 and 2030. That’s one reason why Jiang wants Nodar to repurpose its Hammerhead stereo camera ranging technology to work with park assist technologies.

“I feel that folks that are using these systems today wish they could be better,” Jiang said. “I think that most of us using them, maybe the car can park itself one out of two times.”

While Nodar uses automotive-grade CMOS cameras with Hammerhead, a park assist software version would instead utilize a vehicle’s built-in surround-view cameras as stereo pairs to estimate 3D point clouds. The software required to use these untethered, independently mounted cameras in this way would require three main features, Jiang said: advanced calibration (which Nodar writes and can calibrate the cameras frame-to-frame), perspective distortion and the ability to gather information from low-texture environments like asphalt.

Jiang showed a video of two wide-baseline, five-megapixel cameras generating 50 million points per second. “A nicely reconstructed point cloud makes it a lot easier to do path routing with object detection and understanding generally what’s around the vehicle,” he said. “Stereovision algorithms in the past have not been able to generate such clean point clouds.”

Testing in progress

Nodar’s software would use a car’s built-in cameras as five stereo pairs to create the ability to create detailed, close-up point clouds with 360-degree coverage around a vehicle. (Nodar)

Nodar has been working with an unnamed OEM to pair up common cameras around a vehicle, cameras with either 180- or 120-degree fields of view. The cameras are used as five stereo pairs that together create a 360-degree field of view around the car: two pairs that use the front-facing camera and one from each side mirror, one pair on each side using a camera in the B-pillar and the corresponding side mirror, and the last pair in the back that uses a camera on the rear roof along with the lower-mounted back-up camera.

“You don’t need to have horizontal cameras; you don’t even need to have them in the same plane,” Jiang said. “You could have these sort of arbitrarily mounted, and those five zones are put together to generate this particular point cloud, showing a nice reconstruction of the asphalt with the lane lines and the cars around it. And it goes out to about 12 meters (39 ft) of range.” Current ultrasound sensors used with park assist systems are typically limited to 2 to 4 meters (6.5-13 ft), he said.

When aggregating data from cameras running at 25 frames per second, the result is an aggregated point cloud that shows where the car can park. Jiang said the system has submillimeter accuracy within 4 meters and “millimeter-ish” accuracy out to 12 meters.

“With this kind of resolution, this kind of density, it’s a super easy problem,” Jiang said. “I can see very clearly where everything is, and I know what’s standing above the ground. This, hopefully, will make it very easy for the path planner to decide where to park.”

Nodar’s untethered camera solution would work with Level 2 park assist systems that rely on a human driver paying attention and could be used with Level 3 and above automation when paired with redundant sensor packages, like camera-plus-radar or camera-plus-ultrasound or something else.

An example of Nodar’s software-based 3D-parking reconstruction using video images aggregated over multiple frames. (Nodar)

Nodar, which stands for “native optical distance and ranging,” was founded in 2018. The company has raised $14.5 million from groups like New Enterprise Associates and Rapsody Venture Partners. At AutoSens USA, Jiang said the company will soon open an office in Germany.

Using untethered cameras to define park assist capabilities would also eliminate another complaint people have about current systems. Many of today’s systems use eight to 12 ultrasonic sensors that often get “dirt drips” around them, Jiang said, something customers do not like. Nodar’s system is also more reliable than monocameras that need to be moving to get their Structure-from-Motion (SfM) capabilities to work and are often inaccurate, he said.

The other main option for future park assist systems is lidar, which does well with distance but is “problematic in terms of bill of materials and costs,” Jiang said. Radar is useful for distance measurements as well, but since park assist needs to be able to see lines painted on the ground, it has limits and would require vision cameras to be part of the package.