Realtime 3D model construction with Microsoft Kinect and an NVIDIA Kepler laptop GPU Paul Caheny MSc in HPC 2011/2012 Project Preparation Presentation.

Slides:



Advertisements
Similar presentations
How it works? Actor & green screen 3D virtual background Complete video Camera position Must be very accurate ISLT Broadcaster.
Advertisements

CSE473/573 – Stereo and Multiple View Geometry
SE263 Video Analytics Course Project Initial Report Presented by M. Aravind Krishnan, SERC, IISc X. Mei and H. Ling, ICCV’09.
Results/Conclusions: In computer graphics, AR is achieved by the alignment of the virtual camera with the actual camera and the virtual object with the.
T1.1- Analysis of acceleration opportunities and virtualization requirements in industrial applications Bologna, April 2012 UNIBO.
KinectFusion: Real-Time Dense Surface Mapping and Tracking
Real-time, low-resource corridor reconstruction using a single consumer grade RGB camera is a powerful tool for allowing a fast, inexpensive solution to.
Structured Light principles Figure from M. Levoy, Stanford Computer Graphics Lab.
--- some recent progress Bo Fu University of Kentucky.
Exploiting Graphics Processors for High- performance IP Lookup in Software Routers Author: Jin Zhao, Xinya Zhang, Xin Wang, Yangdong Deng, Xiaoming Fu.
GPGPU Introduction Alan Gray EPCC The University of Edinburgh.
Vision Sensing. Multi-View Stereo for Community Photo Collections Michael Goesele, et al, ICCV 2007 Venus de Milo.
Po-Hsiang Chen Advisor: Sheng-Jyh Wang 2/13/2012.
Stereo Many slides adapted from Steve Seitz. Binocular stereo Given a calibrated binocular stereo pair, fuse it to produce a depth image Where does the.
A Modified EM Algorithm for Hand Gesture Segmentation in RGB-D Data 2014 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) July 6-11, 2014, Beijing,
ECE 562 Computer Architecture and Design Project: Improving Feature Extraction Using SIFT on GPU Rodrigo Savage, Wo-Tak Wu.
Dana Cobzas-PhD thesis Image-Based Models with Applications in Robot Navigation Dana Cobzas Supervisor: Hong Zhang.
Laser Scan Matching in Polar Coordinates with Application to SLAM
Registration of two scanned range images using k-d tree accelerated ICP algorithm By Xiaodong Yan Dec
Motion Detection And Analysis Michael Knowles Tuesday 13 th January 2004.
SIGGRAPH Course 30: Performance-Driven Facial Animation Section: Markerless Face Capture and Automatic Model Construction Part 2: Li Zhang, Columbia University.
Efficient Variants of the ICP Algorithm
A Laser Range Scanner Designed for Minimum Calibration Complexity James Davis, Xing Chen Stanford Computer Graphics Laboratory 3D Digital Imaging and Modeling.
Multi-view stereo Many slides adapted from S. Seitz.
3D full object reconstruction from kinect Yoni Choukroun Elie Semmel Advisor: Yonathan Afflalo.
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 20, NO. 11, NOVEMBER 2011 Qian Zhang, King Ngi Ngan Department of Electronic Engineering, the Chinese university.
Review: Binocular stereo If necessary, rectify the two stereo images to transform epipolar lines into scanlines For each pixel x in the first image Find.
BY: ALI AJORIAN ISFAHAN UNIVERSITY OF TECHNOLOGY 2012 GPU Architecture 1.
Zhengyou Zhang Microsoft Research Digital Object Identifier: /MMUL Publication Year: 2012, Page(s): Professor: Yih-Ran Sheu Student.
KinectFusion : Real-Time Dense Surface Mapping and Tracking IEEE International Symposium on Mixed and Augmented Reality 2011 Science and Technology Proceedings.
Virtual Mirror for Fashion Retailing
MESA LAB Multi-view image stitching Guimei Zhang MESA LAB MESA (Mechatronics, Embedded Systems and Automation) LAB School of Engineering, University of.
Active Pursuit Tracking in a Projector-Camera System with Application to Augmented Reality Shilpi Gupta and Christopher Jaynes University of Kentucky.
Tracking with CACTuS on Jetson Running a Bayesian multi object tracker on a low power, embedded system School of Information Technology & Mathematical.
Fast Support Vector Machine Training and Classification on Graphics Processors Bryan Catanzaro Narayanan Sundaram Kurt Keutzer Parallel Computing Laboratory,
Tracking with CACTuS on Jetson Running a Bayesian multi object tracker on an embedded system School of Information Technology & Mathematical Sciences September.
Stereo Many slides adapted from Steve Seitz.
A Frequency-Domain Approach to Registration Estimation in 3-D Space Phillip Curtis Pierre Payeur Vision, Imaging, Video and Autonomous Systems Research.
Stereo Many slides adapted from Steve Seitz. Binocular stereo Given a calibrated binocular stereo pair, fuse it to produce a depth image image 1image.
Asian Institute of Technology
Lec 22: Stereo CS4670 / 5670: Computer Vision Kavita Bala.
21 June 2009Robust Feature Matching in 2.3μs1 Simon Taylor Edward Rosten Tom Drummond University of Cambridge.
Tutorial Visual Perception Towards Computer Vision
EFFICIENT VARIANTS OF THE ICP ALGORITHM
Based on the success of image extraction/interpretation technology and advances in control theory, more recent research has focused on the use of a monocular.
Spatiotemporal Saliency Map of a Video Sequence in FPGA hardware David Boland Acknowledgements: Professor Peter Cheung Mr Yang Liu.
Computer Science Department, Duke UniversityPhD Defense TalkMay 4, 2005 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Patrick Eibl, Dennis Healy, Nikos P.
COMP24111: Machine Learning Ensemble Models Gavin Brown
CONTENT FOCUS FOCUS INTRODUCTION INTRODUCTION COMPONENTS COMPONENTS TYPES OF GESTURES TYPES OF GESTURES ADVANTAGES ADVANTAGES CHALLENGES CHALLENGES REFERENCE.
Using Adaptive Tracking To Classify And Monitor Activities In A Site W.E.L. Grimson, C. Stauffer, R. Romano, L. Lee.
RGB-D Images and Applications
Cross correlators are a highly computationally intensive part of any radio interferometer. Although the number of operations per data sample is small,
Registration and Alignment Speaker: Liuyu
MASKS © 2004 Invitation to 3D vision. MASKS © 2004 Invitation to 3D vision Lecture 1 Overview and Introduction.
Presenter: Jae Sung Park
Che-An Wu Background substitution. Background Substitution AlphaMa p Trimap Depth Map Extract the foreground object and put into another background Objective.
Design and Calibration of a Multi-View TOF Sensor Fusion System Young Min Kim, Derek Chan, Christian Theobalt, Sebastian Thrun Stanford University.
Scalable Real-time Volumetric Surface Reconstruction
A Novel 2D-to-3D Conversion System Using Edge Information
제 5 장 스테레오.
CS4670 / 5670: Computer Vision Kavita Bala Lec 27: Stereo.
Enabling machine learning in embedded systems
Calibration of Multiple Kinect Depth Sensors for Full Surface Model Reconstruction 2016 the first International Workshop on Pattern Recognition (IWPR 2016)
RGBD Camera Integration into CamC Computer Integrated Surgery II Spring, 2015 Han Xiao, under the auspices of Professor Nassir Navab, Bernhard Fuerst and.
Real Time Dense 3D Reconstructions: KinectFusion (2011) and Fusion4D (2016) Eleanor Tursman.
CAPTURING OF MOVEMENT DURING MUSIC PERFORMANCE
Coding Approaches for End-to-End 3D TV Systems
Combining Geometric- and View-Based Approaches for Articulated Pose Estimation David Demirdjian MIT Computer Science and Artificial Intelligence Laboratory.
3D Scan Alignment Using ICP
--- Range Image Registration
Presentation transcript:

Realtime 3D model construction with Microsoft Kinect and an NVIDIA Kepler laptop GPU Paul Caheny MSc in HPC 2011/2012 Project Preparation Presentation.

Today's Presentation Project motivation. Introduction to Microsoft Kinect and depth maps. A taster on algorithms for constructing 3D models from depth maps. Particular goals and constraints for this dissertation. Paul Caheny 2012

Project Motivation Proposal from Holoxica Ltd. Most holograms produced from 3D models which are computer generated ab initio. Some customers already posses high quality 3D models of real world objects. Commercial scanning solutions prohibitively expensive for Holoxica's purposes. A cheaper, good quality, portable 3D scanning & model construction system could make holograms a more viable and attractive proposition for customers. Paul Caheny 2012

Microsoft Kinect & Depth Maps Depth maps have been the subject of much research for 3D model construction over the past 20 years. The Kinect provides unprecedented quality to cost ratio as a depth sensor. Video Camera, IR Projector, IR Camera IR Projector/Camera constitute a depth sensor providing a 640 x 480 depth map at 30Hz frame rate. Paul Caheny 2012

3D Modelling from Depth Maps Put simply – fusing multiple viewpoints of the scene or object into a single unified 3D model. Early approaches involved highly calibrated motion of the object plus simple surface construction techniques (e.g. directly meshing the depth data points). Current state of the art replaces calibrated motion with software tracking & high quality 3D model construction techniques. Two techniques: Iterative Closest Point (ICP) algorithm for object tracking and Volumetric Integration for high quality model generation. Paul Caheny 2012

Iterative Closest Point (ICP) A method for the alignment of 3D surfaces. Introduced independently by Besl & McKay at GM and Chen & Medioni at Uni. Southern California in early '90s. If corresponding points on surface in two views are known, trivial to compute exact transform which aligns surfaces. But we don't know corresponding points on surfaces in distinct depth maps. Make a heuristic guess of corresponding points, compute transform with selected points, achieving closer alignment. Reselect points heuristically on transformed surfaces, iterate until close enough. Paul Caheny 2012

Volumetric Integration A technique for fusing data from multiple aligned depth maps. ICP alignment & depth sensors have inherent margin of error / measurement uncertainty. Results in noisy data points following ICP alignment. Reduce noise & improve result by combining multiple views using heuristics which minimises noise. Curless & Levoy, Stanford '96 Paul Caheny 2012

What's New? ICP and Volumetric Integration introduced in the '90s. Both techniques have have been subject of much refinement since. In the 90s workflow looked like: Scan -> ICP Batch process -> Volumetric Integration Batch process -> Finished 3D Model. Mid 2000s saw systems with realtime ICP phase plus batch Volumetric Integration phase to create finished model saw publication of research by Augmented Reality Group at Microsoft Research Cambridge demonstrating a realtime, synchronous ICP and Volumetric Integration system called KinectFusion running at Kinect full frame rate of 30Hz. Paul Caheny 2012

My Dissertation Focus on small object scanning and Laptop GPU. Implement the algorithms from scratch using CUDA. Plan to use NVIDIA's next generation Kepler architecture, recently launched for consumer desktop & laptop market. Laptop GPU TDP ~35W versus Tesla 2090 TDP of 225W – Fewer Cores, Less Memory, Less Bandwidth, Lower Clock Speed. Paul Caheny 2012

Wrap Up References: R. Newcombe et al. KinectFusion: Real-time dense surface mapping and tracking. In Proc. 10th IEEE Int. Symp. on Mixed and Augmented Reality, B. Curless and M. Levoy. A volumetric method for building complex models from range images. ACM Trans. on Graphics, P. Besl and N. McKay. A method for registration of 3D shapes. IEEE Trans. on Pattern Analysis & Machine Intelligence, 14:239–256, Y. Chen and G. Medioni. Object modeling by registration of multiple range images. Image and Vision Computing (IVC), 10(3):145–155, Paul Caheny 2012