Hammerhead sharks have exceptional depth perception because of their widely set eyes. NODAR employs wide-set cameras in a similar fashion to calculate depth with exceptional precision. Whether tracking prey in 3 dimensions through the ocean or navigating a complex and changing roadway, a wide-baseline vision system will outperform other types of 3D sensing.
How does it work?
There are three classes of depth cues that are used to derive 3D structure from 2D images.
Triangulation
Based on measuring the parallax of an object from two locations (e.g., from two eyes). The longer the distance between eyes (cameras), the better the accuracy and range.
Perspective
Requires assumptions about object shape, which often leads to incorrect depth estimates when these assumptions are not met.
Shading
Requires assumptions about lighting.
Hammerhead sharks rely primarily on triangulation because it is precise and the other cues are often lost in the featureless ocean. NODAR also employs triangulation and widely positioned cameras to calculate depth at long-range, with exceptional precision.
Hammerhead™ Vision
Like the Hammerhead shark, NODAR’s patent-pending Hammerhead™ technology enables cameras to be separated by large distances and in independently mounted configurations.
This has the unique advantage of letting vehicle manufacturers integrate the cameras in novel locations, such as the rearview mirrors or headlights.
For the first time ever, nearly any obstacle - from a brick to an overturned truck - can be detected hundreds of meters in advance and with exceptional accuracy, giving the vehicle perception system sufficient time for evasive action.
NODAR Hammerhead™ introduces algorithmic innovations that allow camera arrays to be placed independently meters apart without requiring a mechanical stiffener or manual calibration.
Wide baseline between cameras allows for long-range operation (up to 1,000 meters)
System is robust against manufacturing tolerances, even at +/- 3 degrees
Actively compensates for camera perturbations due to road shock, vibration and temperature
A closer look at the NODAR sensor placement.
Hammerhead™ Benefits
By eliminating camera mounting constraints and supporting wide-baseline configurations, NODAR Hammerhead™ brings affordable, high-performance, long-range 3D sensing to the ADAS and AV markets, as well as other verticals where performance vs. price ratio must be very high.
Unconstrained Mounting
Cameras no longer need to be rigidly mounted to a PCB. In fact, they can be mounted in a variety of orientations e.g. offset or vertically.
Automatic Calibration
Road shock, vibration, accidents and temperature fluctuations are all accounted for without an accelerometer or gyroscope.
Low Cost
Using cameras rather than high-bandwith opto-electronics, Hammerhead™ hits price points 10X lower than comparable LiDAR systems.
Efficient Computation
Automotive-grade computing with high-performance stereo camera reconstruction.
Long-range
Wide-baseline camera placement enables long-range (up to 1,000m) highly accurate 3D sensing and the ability to detect small objects at a distance
Nighttime & Bad Weather Performance
Hammerhead™ performs better than LiDAR in bad weather. Dual perspectives from multiple cameras allow Hammerhead™ to work in inclement weather and low light settings.
Industry Standard Hardware
Leveraging advances in CMOS sensors, purpose-built silicon, and GPU technology, NODAR’s platform runs entirely on industry standard hardware.
Long Lifetime
Hammerhead™ has a lifetime of 15 years - ~10x as long as mechanical LiDAR.
How does NODAR compare?
Mono Camera AI
Leverages incoming visual cues to identify limited set of known objects. Infers depth in place of actual distance information.
Drawbacks:
Limited universe of known objects
Ambiguity (e.g. child vs adult)
Frequent failure modes: photos on billboards, buses, trucks
High computational resources
Mono Camera Structure from Motion
Estimates 3D structures from 2D image sequences coupled with the vehicles motion (GPS, odometer).Drawbacks:
3D sensing lag because SfM must wait for motion and image sequence
Limited range where it’s needed most - highway driving at high speeds
High computational resources
Lidar
Illuminates scenes with a scanning laser and uses the reflected signals to determine depth.Drawbacks:
High cost
Limited durability
Limited performance in bad weather
Low resolution
Interference
The NODAR Advantage
Multiple camera angles enables physical measurements to all objects, known and unknown, eliminating inference ambiguity.
Instantaneous measurements per frame at up to 30 FPS and ultra long-range sensing enabled by wide baseline between cameras.
Camera-based system is >15X the area coverage rate with the reliability, robustness, and low price of high volume solid-state cameras.