Hammerhead is NODAR’s patented stereo vision software, delivering real-time, high-resolution 3D mapping for applications where safety, reliability, and performance are critical. With advanced auto-calibration, robust stereo matching, and computational efficiency, Hammerhead excels in diverse environments and challenging conditions, like harsh weather, low light, and long-range object detection.
Patented Auto-Calibration
Natural Scenes
Hammerhead uses natural scenes to continually calibrate the stereo system, eliminating the need for calibration targets, checkerboards, or calibration test equipment. From factory to everyday use, online calibration is always running, so no trips to the service center are needed.
Safe
Safety depends on reliability. Blazing fast algorithms keep Hammerhead's untethered cameras perfectly aligned, even in the face of cobblestone roads, truck vibrations, and challenging off-road conditions.
Long Range
1000+ meter ranges require excellent rectification of images with sub-pixel reprojection errors. Achieving fast and accurate calibration with limited computational resources is our specialty.
Advanced Stereo Matching
Bad Weather
Thanks to a robust matching algorithm, our stereo matcher works in heavy rain, dense fog, and dusty conditions with only minor accuracy degradations.
Low Light
Precise online calibration and proprietary stereo matching algorithms enable Hammerhead to handle blur and low light exceptionally well - better than other camera-based sensors and 2x better than lidar.
Computationally Efficient
Hammerhead strikes a perfect balance between accuracy and speed for automotive applications. Optimized in partnership with Nvidia for the DRIVE Orin automotive-grade SoC, Hammerhead calibrates >5MP cameras in realtime, generating 3D pointclouds every frame without downsampling (common for neural networks). Hammerhead can be ported to other platforms.
Auto-Calibration
Stereo cameras measure distance by triangulation. Increasing the distance between the cameras, amplifies the measured angles, and improves the range estimate. 10x baseline increases range by 10x, making stereo vision a long-range option that rivals LiDAR.
But large baselines are also more sensitive to shock, vibration, and environmental perturbations, making it impossible to manufacture stereo cameras with baselines greater than a few 10’s of centimeters. Until now. To harness the power of ultrawide baselines, NODAR developed online autocalibration software to overcome mechanical mounting limitations and allow mounted pairs of independent monocular cameras almost anywhere, yielding exquisite long-range 3D pointclouds.
Frame-by-Frame Calibration
Bumpy roads, engine noise, or even shifts due to temperature can mis-align cameras and ruin the depth measurements of a stereo system. This problem is amplified with greater distance between the cameras, with higher resolution imagers, and at long range. To maintain camera calibration and the safety of the system, Hammerhead calibrates the cameras every frame, in realtime.
The sequence above shows Hammerhead's instantaneous recovery following a severe perturbation to the cameras.
How does Hammerhead compare?
Hammerhead outperforms other 3D sensors in terms of range and in harsh conditions. Ultra-wide baseline stereo sensing combined with high-resolution cameras and Hammerhead’s precise calibration enable accurate depth measurements even in the face of blur and low light.
With stereo vision, performance is guided by baseline and camera resolution. Hammerhead’s ultra-wide baseline and high-resolution benefit the system at short and long ranges, improving safety and expanding capabilities.
Comparison of weather degradation between LiDAR and Hammerhead
Measuring the ratio of valid vs. total possible depth measurements, NODAR outperforms industry standard lidars by over 2x in poor visibility conditions.
Long Range
Hammerhead's long range is enabled by several factors: sub-pixel cameras calibration, high-resolution imagers, and wide-baselines. Combined, these capabilities allow Hammerhead to "see" as far as 1000 meters.
Forward Detection Range
Hammerhead 500 m
LiDAR 50 m
Hammerhead in Action
Hammerhead outputs available via API include: raw RGB data (upper-right), depthmap (upper-center), BEV (right side), and pointcloud (bottom), and confidence map (not shown)
Hammerhead Architecture
Specifications
Calibration
Continuous (every frame)
Baseline
Tested up to 3m
Frame rate
Up to 30 FPS
Max capacity
50 million pixels/second ¹
Accuracy
0.2% @150m ²
Maximum range
Tested to 1,000m
Precision
Able to detect 10cm object at 150m ³
Camera alignment tolerance
+/-3°
Stereo matching algorithm
Deterministic signal processing algorithm
Vibration tolerance
0.1° per frame
1. Using NVIDIA Orin AGX
2. Using 5.4MP Sony IMX490 cameras, 30° FOV, 1.2m baseline
3. Dependent upon resolution and frame rate