Selected Projects

Exvision Gesture System (EGS) Motion

A gesture camera powered by Movidius Myriad2 Vision Processing Unit which can be used to recognize simple gestures. By analyzing motions at 120 FPS, simple gesture motions can be recognized with low latency and enabling gesture recognition on devices can be as simple as connecting this camera using USB interface. This is the first product I developed at Exvision where I developed the core algorithm from scratch and submitted a patent for it.

Aerial Imaging by Retro-Reflection (AIRR) Tablet

Have you ever seen movies where a screen appears to be floating in the air? This project is exactly trying to turn that into a reality. During my study at The University of Tokyo at Ishikawa-lab, I created a prototype of a floating display based on Aerial Imaging by Retro reflection technique developed by Assoc. Prof. Hirotsugu Yamamoto []. I created a 3D hand high-speed (500 FPS) hand tracking system to make the display interactive. The low latency and high FPS makes the interaction very natural.

High Speed Gesture Recognition System for zSpace 3D

Similar to the Aerial Imaging by Retro-Reflection (AIRR) Tablet, this project aims to demonstrate how we can interact with a virtual world using our hands. However, instead of showing a flat screen floating in the air, we show a virtual 3D objects using a novel display system called zSpace. This was a collaboration project with the maker of the display system, zSpace.

High Speed Ball Tracking System

Tracking a high speed moving object where the appearance varies with time is challenging. I developed a machine learning based approach to object tracking which is robust to lighting, background variations and varying object appearance. The system is able to track ball in real-time at 500FPS.

Ball Rotation Estimation System

High speed moving objects may also rotate quickly. This system complements the tracking system by analyzing the rotational motion of the tracked objects based on 3D template. The rotation analysis is accelerated by using GPU. The system is able to analyzing a segment of trajectory to estimate the axis of rotation and angular velocity of the ball.

Cable Defects Detection System

Developed a defect detection system for cable defects taking X-Ray image as input. Currently deployed in production plant in one of biggest cable makers in Japan.

Machine Learning Education Program

How can we teach machine learning to children? I developed a desktop application to teach children how we can teach machines. Using a real-robot, children can teach the robot to perform simple task such as following the owner and simple navigation rules. This work is the result of collaboration with NHK Enterprise, Japan.