Ray-transfer functions for camera simulation of 3D scenes with hidden lens design

Area of Science:

• Optics and Photonics
• Computer Graphics
• Machine Learning

Background:

• Image sensor simulation and physically based ray tracing are crucial for designing novel imaging systems.
• Synthesizing physically accurate, labeled images aids machine learning applications.
• Simulating lens optics precisely is challenging due to proprietary lens designs.

Purpose of the Study:

• To present a pragmatic solution for simulating lens optics accurately while preserving intellectual property.
• To enable precise soft prototyping of imaging systems.
• To facilitate the synthesis of high-fidelity training data for machine learning.

Main Methods:

• Developed a black box lens model using Zemax software.
• Created software to construct a polynomial ray transfer function characterizing lens behavior.
• Implemented the ray-transfer calculation as a camera model in PBRT (Physically Based Rendering Toolkit).

Main Results:

• The polynomial ray transfer function accurately models lens optical properties.
• PBRT ray-transfer calculations were validated against Zemax lens calculations.
• Edge spread functions and relative illumination were accurately reproduced.

Conclusions:

• The black box lens model offers a viable method for accurate optical simulation in imaging system design.
• This approach balances the need for precise simulation with the protection of intellectual property.
• The validated method supports advanced imaging system development and machine learning applications.