Projects

  1. Home
  2. Projects

Automatic Presbyopia Correction Glasses

By integrating focus-tunable lenses, miniature cameras for pupil tracking, and real-time vergence control, we developed a wearable system that dynamically adjusts optical power to restore clear vision at all distances. This approach enables seamless, automatic focus adjustment for users with presbyopia, enhancing visual comfort and daily convenience.

Paper (legacy version)

VolRec 4D

VolRec 4D is an advanced algorithm for real-time volumetric reconstruction and processing of large-scale OCT data. By leveraging both spatial and temporal reconstruction techniques over GPUs, VolRec 4D enables the acquisition and visualization of high-resolution OCT volumes at rates exceeding 25 volumes per second, making it ideal for dynamic, high-throughput imaging applications.

Paper

Pupil Tracking GPU

Our GPU-accelerated pupil tracking algorithms leverage CUDA to deliver up to 15× speed improvements over sequential methods, with no loss of accuracy. This enables real-time, high-precision pupil detection and tracking for demanding imaging applications.

Code

Blob Finder C++

Blob Finder C++ is a parallelized C++ algorithm for efficient blob detection, inspired by the blob detection tools in the scikit-image suite. It uses a multi-scale Laplacian of Gaussian (LoG) approach to accurately identify blobs of varying sizes in images, and is optimized for high performance on multi-core CPUs.

Code

Ellipse Fit CUDA

Ellipse Fit CUDA is a GPU-accelerated algorithm for fast and accurate ellipse fitting, utilizing CUDA and its linear algebra library (cuBLAS). It supports efficient batch processing, enabling simultaneous fitting of multiple ellipses for high-throughput image analysis tasks.

Code

DeepPT

DeepPT is a machine learning-based algorithm that enables fast, precise, and energy-efficient pupil tracking. By leveraging advanced neural networks, DeepPT delivers real-time performance and high accuracy while minimizing computational power, making it ideal for resource-constrained and embedded imaging applications.