Guan Ming Lim

Google Scholar  /  Github  /  YouTube

I'm interested in combining research on computer vision, computer graphics and machine learning to understand human motion. Much of my research is on hand and object tracking from images.

The RRIS40 dataset uses anatomical landmarks from marker-based motion capture combined with deep learning techniques to achieve accurate 3D markerless human motion capture.

Real-time, accurate, and robust tracking of a rigid object using image-based methods such as efficient projective point correspondence and precomputed spare viewpoint information.

A low-cost, modular, and wireless pressure sensor array that can generate real-time pressure distribution map for object pose estimation and grasp classification.

Real-time estimation of 3D hand shape and pose from a single RGB image running at over 110 Hz on a GPU or 75 Hz on a CPU.

A camera-based system for markerless hand pose estimation using a mirror-based multi-view setup to eliminate the complexity of synchronizing multiple cameras and reduce the issue of occlusion.

Semantic segmentation of body parts (e.g. hand and arm) and objects from depth image. A Fully Convolutional Neural Network is trained on synthetic data with some level of generalization on real data.

SDF-Net is a simple multilayer perceptron (with a memory footprint of < 10 kB) that is used to model the signed distance function (SDF) of a rigid object for real-time tracking (≈ 1.29 ms per frame on 1 CPU core) of the object using a single depth camera.

Design and source code from Jon Barron's website