Immersive 360° VR Experience

Project notes — Jan 2024 to May 2024

This post summarizes the key technical components of my bachelor's thesis and related work on panoramic video stitching and immersive VR. It highlights the feature-based matching, homography estimation, cylindrical projection to reduce frame skew, camera calibration, and CUDA-accelerated warping.

Key contributions

• Feature-based matching using KNN on SIFT descriptors for robust frame alignment.
• Homography estimation for accurate image registration and warping.
• Cylindrical projection to reduce frame skew and improve stitching quality.
• Camera calibration routine to estimate intrinsic parameters for cylindrical mapping.
• GPU acceleration using CUDA for warping — achieved 1.3x–2x speedup.

Implementation notes

Implemented using OpenGL and OpenCV for low-level image processing and rendering. CUDA kernels were written to parallelize the warping step and reduce per-frame processing time.

← Back to blog