Cloud-Based High-Precision Stereo Depth Estimation

Overview

NODAR GroundTruth is a cloud platform that delivers the most accurate stereo depth maps and 3D point clouds from synchronized image pairs. Designed for precision, it leverages a state-of-the-art deep stereo matching algorithm to generate ground truth-quality 3D data. This solution is ideal for customers requiring highly accurate depth information for applications such as map generation, HD dataset labeling, or training and validation of AI models.

Inputs

• Synchronized Image Pairs
  • Format: TIFF
  • Resolution: Up to 5.4MP for full-scale processing
• Per Frame Calibration Parameters (optional)
  • Required Accuracy: Angular Accuracy of +/- 10 pixels per Relative Camera Rotation Angle
• Initial Calibration
  • Required Accuracy: Angular Accuracy of +/- 3 degrees per Relative Camera Rotation Angle
  • Required inputs:
    • Baseline length
    • Distance to a reference object
  • Please contact [email protected] to assist in calibration

Outputs

• Rectified Images
  • Calibration Accuracy: 0.25 px
• Disparity Maps
  • Average Disparity Accuracy: 0.13 px
  • Disparity Precision: 1/16 px
  • Format: 16bit single-channel TIFF
• 6D Point Clouds
  • Depth Precision: depending on the baseline and camera focal length, see below for sample accuracy curves
  • Format: XYZ-RGB compressed LAZ
• Transformation Parameters
  • Reprojection Matrix Transforming Disparity Maps to Point Clouds
  • Rotation Matrix between the Rectified and Raw Camera Space

Requirements

• Access to Command-Line Interface (e.g. Terminal)
• AWS S3 Data Storage Bucket
• Stereo Camera Hardware Requirements
  • Horizontal geometry with at least 50% overlap between the fields-of-view
  • Narrow and wide camera fields-of-view: $10^\circ-180^\circ$
  • Pinhole and Fisheye camera models
  • Arbitrary baseline length, tested between 5cm and 3m
• Image Timestamp Synchronization
  • Synchronization requirements depend on the camera and object speeds. However, synchronization of $\pm 20\mu s$ is sufficient in a vast majority of scenarios.
  • For a detailed discussion of the synchronization requirements, please refer to
    • Synchronization requirements on static scenes
    • Synchronization requirements on dynamic scenes

Platform & Access

• User Access
  • Command Line Interface
  • REST API (available soon!)