Nodar invents stereo vision calibration for autonomous cars

Stereo vision, which was first used in the late 1800s, has now entered the autonomous driving space. Nodar, a Boston-based company that develops software to align camera measurements by software, is currently developing an algorithm. This is a de...

Stereo vision, which was first used in the late 1800s, has now entered the autonomous driving space. Nodar, a Boston-based company that develops software to align camera measurements, is developing an algorithm. Nodar's software, which is faster than traditional stereo vision, can maintain alignment by signal processing.

Brad Rosen, co-founder at Nodar and COO of Stereo Vision Systems, stated that "Stereo vision system are very sensitive to relative alignment between cameras".He said that even a misalignment of one hundredth of a degree could lead to an incorrect measurement. Legacy stereo vision requires a rigid beam to secure the cameras. The distance these systems can see is only about 50 meters, however, because the cameras must be placed very close together on a beam.

Rosen says that Nodar's Hammerhead stereovision system, named after the Hammerhead shark, has many advantages over other lidar systems for autonomous driving. Cameras are cheaper, have better resolution and last longer than lidar. Cameras are therefore suitable for mass-produced vehicles. Hammerhead allows the cameras to be mounted on the vehicle independently and far apart. This allows the system to see far ahead (upto 1,000 meters). Hammerhead can also be used with any type of camera.

Nodar's online calibration takes only 3 TOPS processing power. Calibration occurs every frame in realtime. The system was demonstrated at CES 2023 in January and can create between 10-20 frames per second with up 5 million pixels per frame.

Rosen stated that software-enabled processing would not have been possible ten years ago without ultra-fast GPU processors. He also stated that the public research by Daimler & Others showed that stereo vision performed best in bad weather conditions than other optical systems (see Note1 for reference).

Nodar unveiled Hammerhead DevKit at CES this year. It uses a Lucid Triton 5.4MP cam and NVIDIA Orin Jetson process. Rosen stated that the company has been working with European OEMs and suppliers for many years to deliver the product in the automotive industry. They hope to announce new partners this year. Most likely, the Hammerhead system will be used first on L4 autonomous trucks.

The company has conducted many proof-of concepts (POCs) in other markets that could lead to operation autonomy: air taxis and heavy equipment, construction cranes as well as autonomous ferries, personal boats, and autonomous farming.

Rosen stated that the Nodar system's value is in its software. There are still requirements for hardware, but automotive-grade cameras can be used and GPUs can carry it.

Nodar's stereo vision with a wide baseline allows it to see up to 1,000m ahead. It is crucial for autonomous trucking fleets that they can detect obstacles as far as possible on the highway. Rosen stated that autonomous trucking fleets must be able to spot dangerous objects from a distance. Truck crashes can often prove fatal and must be avoided.

The system can detect small objects as small 10 cm at 150m away, or a motorcycle that has been flipped at 350m on the highway, as long as they are within the camera's overlapping fields of vision.