Introduction to Computer Vision (CS4670), Fall 2010
Final Project
Assigned: Sunday, October 24
Proposal Due: Friday,
October 29 by 11:59pm (via CMS)
Status Report: Friday,
November 19, 11:59pm
Final Presentation Slides
Due: Sunday, December 12, 7:59pm
Final Presentation: Monday, December 13, 9-11:30am, *** Location: ***
Phillips 203
Writeup/Code Due: Tuesday, December 14,
11:59pm
Quick Links:
Synopsis and Guidlines
Requirements
Resources
Final Project Ideas!
Synopsis and Guidelines
The final project is an open-ended project in mobile computer
vision. In teams of two or three, you will come up with a project
idea (some ideas to get you started are described below), then
implement it on a Nokia N900. Your project must involve some kind of
demo that you can show in class, either through a live demo or a video
that you have created of the system in action. The project could be a
new research project, or a reimplementation of an existing system, but
must involve a non-trivial implementation of an computer vision
algorithm or system. Teams of three will be expected to devise and
implement a more ambitious project than teams of two.
The goal of the project is (1) to learn more about a subfield of
computer vision, and (2) get more hands-on experience with computer
vision on mobile devices.
How ambitious/difficult should your project be? Each team
member should count on committing at least twice the work as in
Project 2ab.
Accordingly, you won't be able to implement something arbitrarily
ambitious, but please feel free to use your imagination when coming up
with projects, and to implement a prototype system that could
be extended in interesting ways. As part of the project, you can use
any capability that you can think of. For instance, the phone comes
with wireless, a touch screen, accelerometer, GPU, GPS, and other
bells and whistles. You can set up a remote server that listens for
requests from the phone, and runs some vision algorithm on the server.
You can use Google Streetview, or any other existing API on the Web
(as long as you still implement something interesting yourselves).
Requirements
Proposal
Each team will turn in an approximately one-page proposal
describing their project. It should specify:
- Your team members
- Project goals. Be
specific. Describe what the inputs to the system are,
and what the outputs will be.
- Brief description of your
approach. If you are implementing or extending a previous method,
give the reference and web link to the paper.
- Will you be using helper code
(e.g., available online) or will you implement it all yourself?
- Breakdown--what will each
team-member do? Ideally, everyone should do something imaging/vision
related (it's not good for one team member to focus purely on
user-interface, for instance).
- Special equipment that will
be needed. We may be able to help with servers, extra
cameras, etc.
Turn in the proposal
via CMS
by Thursday, October 28 (by 11:59pm).
Status Report
Each team will turn in a one page status report for their project
on Friday, November 19 by 11:59pm. This report should
present your progress to date, including preliminary results, as well
as any problems that you are encountering.
Final Presentation
Each group will give a short (10 minute) PowerPoint presentation on
their project to the class. Details will be announced closer to
the time of the presentation. Your final presentation slides should
be uploaded to CMS.
Your presentation is expected to include some kind of (canned or live)
demo.
Final Writeup
Turn in a web writeup describing your problem and
approach. This writeup should include the following:
- title, team members
- short intro
- related work, with references
to papers, web pages
- technical description
including algorithm
- experimental results
- discussion of results,
strengths/weaknesses, what worked, what didn't
- future work and what you
would do if you had more time
Code
In addition to the
writeup, you will be turning in the code associate with the
project.
Resources
Coming soon...
Final Project Ideas!
Here are several ideas that would make appropriate final
projects. Feel free to choose variations of these or to devise
your own projects that are not on this list. We're happy to meet
with you to discuss any of these (or other) project ideas in more
detail--if you can't make office hours, just email the instructor to
set up a meeting.
- Nokia Goggles Lite. Write an app that can recognize some
limited class of objects, like all book covers or all DVDs, all
artwork, etc. You may need to write a scraper to download large sets
of images from Amazon, for instance, in order to create the database.
You'll probably also need to create a server for this project.
- Computational
Photography App. Computational photography (which we
will talk about in class) uses computation combined with imaging
to create better images (your panorama stitcher in Project 2 is
an example of a computational photography application). A mobile
device---combining a camera with computation---is an ideal
platform for computational photography. Devise and implement
a computational photography app on the N900. Here are some
possibilities:
- HDR (high dynamic range) imaging. Cameras are
limited in the dynamic range (i.e., the range of intensities of
light hitting the sensor) they can capture in a single photo; it
is hard to capture very bright and very dark intensities in the
same photo. However, if we take multiple photos with different
exposures, we can combine them together to produce an HDR (high
dynamic range) image. You can see many examples of HDR images on Flickr.
Write an app for taking multiple photos with different exposures
and combining them on the phone. See a related project from Li
Zhang here.
You can also get inspiration from the iPhone's version of this
app, reviewed here on Ars
Technica.
- 360 panorama capture. Extend your panorama app in
Project 2b to create an entire 360 panorama on the phone (see the
Monster Bells on the Project
2b page). Alternatively, make a real-time panorama capture
app as described in this
project, or as implemented in the N900
QuickPanorama app.
- Flash-no flash. Use the flash in the N900 to capture
flash-no flash pairs then combine them into beautiful images, as
described in this
project.
- Video stabilization. Use optical flow to create a
real-time image stabilizing algorithm.
- Image deblurring. Can you use the phone's
accelerometer to help with image deblurring? See this
project.
- Something else cool. Take a look at the
Frankencamera
project for more ideas for computational photography apps.
- Location-based games. The instructor is working on
PhotoCity, a game for
photographing all of Cornell. Part of this game is an app that
runs on a mobile device. Build an N900 version of this app. For
more details, talk to the instructor.
- Location
recognition and augmented reality. Create a Cornell
campus app that recognizes what building you are in front of by
taking a photo and comparing it to a large database of Cornell
images. If you are interested in this project, please talk to
the instructor (who has tens of thousands of images of Cornell
that could be used as a database). Use this in a campus tour,
which displays additional information on top of the photo, such
as the name of the building and a link to Wikipedia.
- Misc. Augmented Reality or Recognition App. There are
many things you could do here. Take a picture of an airplane flying
overhead, and automatically highlight the plane, along with the
flight number, by connecting with an online flight database (and
estimating the rough location and orientation of the phone). You
could write a barcode scanner, and display useful information on an
image of a product. You could write a vision-based Sudoku-capture
app for taking a photo of a Sudoku puzzle and converting it to a
digital version. And so on.
- Digital object insertion. Build an app to track the pose
of a camera, and insert a digital 3D object into the real scene (as
viewed through the phone. This makes the phone into an interface for
viewing a virtual 3D object by simply walking around it.
- Face
recognition. Write an app that can take an photo of a person
and recognize them and display their name.
- Artistic image
filtering. Create an image filtering app that in
real-time applies an interesting (non-linear) filter to the
stream of images. For instance, you could apply a cartooning
effect, as in this project on
real-time video abstraction.
- Vision-based user interface. Write an app that will
implement a phone UI based on computer vision (this is most useful for
a phone with a front-facing camera, e.g. an iPhone 4, but we can at
least prototype one with the N900). If might track features on your
face, for instance, or recognize gestures, in order to activate
certain UI commands (e.g., ``raise left eyebrow'' might push the ``1''
button on the dialpad---you can probably think of much more
useful ideas). This could be used as an interface for impaired users,
or as an interface to a new game (imagine using your face to control a
game character).
- Stereo/structure from motion. Use the camera to capture
two images, then run stereo on them to produce an image with depth
map. You'll first need to estimate the F-matrix between the two images.
- Autofocus. Create a camera app that implementes autofocus
by recognizing faces in the image.