GSoC 2025 Proposal: OpenCV Multi-Camera Calibration Python Toolbox

[Harshith Mohan Kumar]

Hi Jean-Yves Bouguet, Gary Bradski, Alexander Smorkalov, Maksym Ivashechkin, and Jia Wu,

I’m a graduate researcher at UC Riverside’s Computational Optics and Video Computing Group, interested to contribute to OpenCV’s multi-camera calibration and computational photography projects for GSoC 2025.

Recently, I built a multi-camera calibration pipeline for LWIR-RGB systems using synchronized stereo pairs. This addresses Kalibr’s abstraction limitations and ROS dependency issues.

My calibration approach extends OpenCV’s cv2.findHomography for synchronized multi/stereo capture systems through iterative refinement using RANSAC, offering a lightweight alternative to Kalibr. It also allowed for flexibility in tuning image processing parameters for more refined chessboard corner point detections (using techniques like clahe, sharpening, etc without injecting significant error in corner point detection).

While the code remains private for conference review, I plan to open-source it post-publication and am happy to discuss implementation details privately.

Jean-Yves Bouguet and Gary Bradski: I’d appreciate your insights on integrating this work into OpenCV’s calibration modules. Are there specific points in multi-sensor calibration I could address this summer?

I’m also open to the other multi-camera calibration proposals and I’ve taken an interest in the computational photography algorithms proposal. I’m quite eager to contribute to an open-source project this summer!

Past GSoC experience: Red Hen Lab 2022 (project link) — built a multi-modal TV segmentation tool.

I've linked a short video showcasing the results of my calibration script :)

Best regards,
Harshith Mohan Kumar
Email: hmoh...@ucr.edu
Portfolio: https://harsh188.github.io/