Dimensioning Cuboid and Cylindrical Objects Using Only Noisy and Partially Observed Time-of-Flight Data

Area of Science:

• Robotics and Automation
• Computer Vision
• Metrology

Background:

• Time-of-Flight (ToF) sensors offer depth sensing but face challenges like low resolution, noise, and multipath interference.
• These limitations distort object shape and size, hindering accurate dimensioning applications.

Purpose of the Study:

• To apply a superquadric fitting framework for accurate dimensioning of cuboid and cylindrical objects using ToF sensor data.
• To evaluate the framework's performance across various object orientations, ground surfaces, and fitting techniques.

Main Methods:

• Utilized point cloud data generated from a ToF sensor.
• Applied a superquadric fitting framework to model and dimension cuboid and cylindrical objects.
• Investigated bounding and mirroring techniques for model fitting.

Main Results:

• Achieved average dimensioning errors below 1 cm for objects up to 30 cm (cuboid) and 20 cm (cylinder) at 1.5 m distance.
• Demonstrated absolute dimensioning errors of 4%-9% for cuboids and 2.97%-6.61% for horizontal cylinders using the bounding technique.
• Showcased orientation-dependent performance, with vertical cylinders exhibiting higher errors (8.01%-13.13%).

Conclusions:

• The superquadric fitting framework effectively overcomes ToF sensor limitations for accurate object dimensioning.
• The method provides quantifiable accuracy across different object shapes, orientations, and surface conditions.