Paper Reading #1: KinectFusion, Real-time 3D Reconstruction and Interaction Using a Moving Depth Camera

Reference Information

Shahram Izadi, David Kim, et al. (2011) "KinectFusion: Real-time 3D Reconstruction and Interaction Using a Moving Depth Camera." KinectFusion. Bibliometrics, 16 Oct. 2011. Web. 30 Aug. 2012.

Authors

Shahram Izadi - Coleads the interactive 3D technologies group at MSR Cambridge..

David Kim - Member of Digital interaction Group at Culture Lab in MSR Cambridge

Otmar Hilliges - Post doc researcher at Sensors and Devices Group at Microsft Research.

David Molyneaux - Intern with the SenDev team in MSRC studying software engineering.

Richard Newcombe - Works at MSR Cambridge in the Computer Vision Group.

Pushmeet Kohli - Researcher for Microsoft researching intelligent machines.

Jamie Shotton - Researcher at MSR Cambridge in the Algorithms Computer Vision Group.

Steve Hodges - Leads Sensors and Devices at MSR Cambridge on accessories.

Dustin Freeman - Working on a Ph.D. in Computer Science at University of Toronto.

Andrew Davison - Leads Robot Vision Research Group and teaches at Imperial College.

Andrew Fitzgibbon - Resraches for MSR Cambridge on Computer Vision and Videos.

Article Summary

Photo Credit: KinectFusion - (http://research.microsoft.com/en-us/projects/surfacerecon)

The article on KinectFusion discusses applications of using real-time hardware to recreate a virtual environment. Microsoft's Kinect, the main piece of hardware involved in this article, has a defect previously mentioned in this article. KinectFusion is an attempt by the author(s) to fix those errors and create a 3D model of the image being recorded from the data. In addition to using the hardware's passive cameras, online images, and tracking info, the authors also develop a algorithm for the hardware that brings environment attributes to play. The previous addition allows for an environment to be effectively controlled and mapped. Because of the limitations of the Kinect hardware, the author(s) had to extend the graphics pipeline to allow for an increase in interactivity.

In the end, the author(s) were able to create a almost real-time virtual environment interaction with a piece of hardware initially designed to only interpolate static data. With the additions they added, they shown it was possible to combine real-world environments with a virtual recreation. Humans and Technology's interaction can be easily seen in this example.

Related Work

When searching for topics related to this paper, I found many results describing what the KinectFusion design does. When analyzing their approach, they all seemed to be intrigued at the concept of being able to create a virtual mapping of an environment in real-time. They described Kinect's flaws just as the authors above had done and told what KinectFusion accomplishes by it's re-design. They talked about other design attempts by other authors but in the end praised KinectFusion for providing a more robust feature set. A Majority of the papers were titled the same as the paper and were featured on Microsoft's Research website. There were even YouTube videos called KinectFusion HQ and Real-Time KinectFusion that explained how the author's concept works.

Evaluation

I feel this work was evaluated unbiasedly and used low-level systematic approaches to redesign the hardware. They first talked about their problem with their current hardware and brought up resolutions. They measured it using quantitative descriptions, describing each piece of hardware inside their design and systematically combining all the parts into one to create a virtual environment. The way they described it was hard to understand, but was worth while.

Discussion

This paper is important because it shows that humans and technology can interact with each other in an almost real-time manner. Innovation on pre-existing concepts shows that there is still much that can be accomplished and technology is truly not at a stand-still.

The authors began by pointing out the flaws of Microsoft's Kinect and began to explain how they tried to solve them. By programming at a level deep enough to extend the GPU pipeline, they allowed for true innovation and helped their product design flourish.

This article shows me that being able to understand technology entirely at a low level is important to truly innovate a product. While hard to understand, I feel the evaluation was appropriate given by the authors.