QR Code Augmented Reality Tracking with Merging on Conventional Marker based Backpropagation Neural Network
QR Code Augmented Reality (QRAR) is an Augmented Reality does not require pre-registration, it has combination ID-encoded and can be used on the public AR application. The results from previous research are 6 DOF tracking method less accurate, require small computation power and unstable marker. We propose merging conventional marker with QR Code, but it will have noise on the QR Code Finder Patter (QRFP) under perspective distortion, so we propose a Backpropagation method to keep detecting the QRFP and the method preceded by feature extraction with low level image processing. The methods we have proposed, achieve accurate 6 DOF, runs at 35.41 fps and stable marker as conventional marker.
Publisher: Intitute of Electronics and Electricals Engineer (IEEE) and Faculty of Computer Science, Universitas Indonesia
Organization: State Islamic University (UIN) Syarif Hidayatullah
Conference Name: International Conference on Advanced Computer Science and Information System (ICACSIS) 2012
Conference Date: Dec 1-2, 2012
Applying Merging Convetional Marker and Backpropagation Neural Network in QR Code Augmented Reality Tracking
Usability of QR Code in Augmented Reality system has been used for digital content accessible publicly. However, QR Code in AR system still has imprecision tracking. In this article we propose merging QR Code within conventional marker and backpropagation neural network (BPNN) algorithm to recognizing QR Code Finder Pattern. The method which our chosen to approaching conventional marker. The result of BPNN testing, QRFP detected in perspective distortion with ID-encoded character length 78, 53 and 32. The result has accuracy of 6DOF ±10.65º pitching, ±15.03° yawing and ±408.07 surging, marker stability has 97.625% and computation time runs at 35.41 fps.
Journal Name: International Journal on Smart Sensing and Intelligent Systems (S2IS)
Publication Date: Dec 15, 2013
Interactive Dialogue Technique Based Computer Vision with Palm Tracking
This paper aims to propose natural interactive technique those using biometrics which do not require direct contact or physical engagement with user’s input device. One of the natural motion that can be used as a tool is palm tracking. A prototype palm tracking made to control mouse pointer (left click) and cursor keys (left and right arrows) use Haar Cascade to detect movement or shifting of each pixels an object in real-time video. The method is using haar-like features which need to be training first to get a decision tree (cascade classifier) as a determining of whether there is a palm object or not in each frame which being processed. The prototype has been made to run in real time.
Publication Date: Jan 11, 2014
Conference Name: International Journal of Information Technology & Computer Science