DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING
UNIVERSITY OF CALIFORNIA, SAN DIEGO

• Fun with Google Satellite Images
• Build detector for structures not appearing on map, e.g., swimming pool, race track, baseball diamond, etc.
• Use data from http://maps.google.com or http://earth.google.com/.
• training data available from wikimapia.org
• birds-eye photos available at live.com
• locate solar panels on houses
• tie in with Zillow and census data, assist in urban planning
• Recognizing Cars
• Build off of Louka Dlagnekov's MS thesis Video-based Car Surveillance: License Plate, Make, and Model Recognition
• improve OCR algorithm
• detection of car manufacturer logos/insignias
• bumper sticker detection
• improve LPR performance using super-resolution
• classify car colors into familiar color categories (silver, red, etc.)
• estimation of 3D structure from 2 different poses
• Analysis of Underwater Images and Video of Coral Reefs
• In collaboration with SIO.
• Color calibration using color charts.
• Detect bleaching in coral.
• Material classification on seafloor.
• Website illustrating project goals.
• Touch-based interactive image manipulation
• Specify image transformations (zoom, rotate, scale, perspective distortion) with multi-finger contact and gestures.
• See demo from Jeff Han at NYU.
• Collaborate with HCI group in CogSci dept.
• 3D Object Scanner (on a Budget)
• Milkscanner V1.5
• requires 1 webcam, 1 tupperware bowl, 3 cups of milk, and 1 custom LEGO rig
• Assistive Technologies for the Visually Impaired
• see the research projects at Smith-Kettlewell Eye Research Institute
• read street signs, bus stop signs, detect crosswalks
• if unknown object detected, upload to internet, make software for volunteers on web to identify (maybe like LabelMe interface)
• develop Remote Sighted Guide (RSG) infrastructure for assisting blind people in grocery shopping, leveraging ideas from peekaboom and espgame.
• Automatic Mosaic Construction
• Build photomosaics without hand-clicking points.
• Learn about Block (and Bundle) Adjustment.
• Implement algorithms by Brown & Lowe and Shum & Szeliski.
• Diagnosis of Skin Diseases
• In collaboration with UCSD Medical School and OGI.
• Distinguish between normal skin and basal cell carcinoma (BCC).
• Use images from polarized light camera as a guide.
• For more information, see http://www.bme.ogi.edu/biomedicaloptics/jacques/polCAM/.
• Handwritten Digit Recognition
• MNIST Dataset of Handwritten Digits
• Virtual Simultaneous Replay
• SimulCam is used with events such as downhill skiing to depict two competitors performing a race at the same time.
• As a project, implement a basic version of this, e.g., using 2-3 Pan-Tilt-Zoom (PTZ) cameras on campus somewhere capturing two people running a race.
• Animal Monitoring
• Multiple Mouse Tracking
• Mouse Behavior Recognition
• Smart Vivarium Project Website
• Mobile Vision Platforms
• Develop mobile computer vision applications on an ARM9 core [www]
• Implement vision algorithms on an ER-1 robot [www]
• Super-Resolution
• How to combine multiple low-resolution images of the same scene to make a composite high-resolution image.
• References: Capel, Zisserman, Irani, Peleg, Milanfar.
• Handwritten Equation Recognition
• See for example MIT's Natural Log project.
• Sea of Cell Phone Cameras
• In a typical classroom, dozens of students have cell phones with cameras. How can images from all of them be used together to solve some interesting problem?
• Object Recognition
• LabelMe: The open annotation tool (at MIT)
• Caltech object recognition databases
• PASCAL Visual Object Class Challenge (this is different from PASCAL the programming language)
• How to segment and/or recognize food on a plate, as a diet assistance tool.
• Human Surveillance
• In static photographs: help a news agency count the number of people at a demonstration.
• In video: count number of people walking through a corridor.
• CAVIAR database for human activity recognition.
• Online Photo Sites
• See for example this project from Berkeley that uses face images and attractiveness ratings from the ``Hot or Not'' website.
• Riya.com does face and text recognition in photo albums. It could be interesting to tie this in with Google or Yahoo image search to find celebrities, political figures, et al.
• Book/CD/DVD Cover Recognition
• Delicious Library manages media collections via barcode scanning with an iSight camera, and grabs cover art from Amazon.
What about recognizing the cover art directly, instead of using the bar code?
• Parking Lot Monitor
• Point one or more webcams at a parking lot.
• Detect vacant parking spaces.
• 3D Photography on Campus
• Use stereo/multiview/video to reconstruct 3D models of sculptures around campus.
• Examples: Bear, Sun God.
• Relevant courses at udub and Caltech. More links here.
• Use data from Mars rover
• Google Mars may be a good resource.
• also the NASA Mars Rovers website.
• Fish Recognition
• Large collection of images available at the Digital Fish Library.
• Facial Image Registration
• Watch the video Noah takes a picture of himself every day for six years
• Use automatic image registration to align the facial features across all the images.
• Pedestrian Detection
• use a car mounted camera to do pedestrian classification
• do experiments with DaimlerChrysler database
• TA grading assistant
• perform OCR on students' handwritten names and ID numbers
• use constraint that names and numbers must be drawn from class roster known ahead of time
• help TAs of lower division CSE classes with a real-world paperwork management problem involving scanned quizzes!
• Back-of-Head Recognition
• How well can you recognize someone who is facing away from you?
• Consider a photograph of a conference room taken from the back: can you recognize individuals from their hair pattern/color, ears, neck, etc.? Can the person's slouch/posture also play a role?
• You can think of this as the extreme case of face recognition with `out of plane' rotation invariance.
• As part of the study, you'd build a database and do a study of human ability to perform this task.
• USPS Mailbox Detector
• Detect blue mailboxes in photos like these: http://www.payphone-project.com/mailboxes/.
• Challenging due to variability in scale and illumination.
• Use object detector such as AdaBoost cascade with color and brightness features.
• 3D Laser Scanning
• DAVID: freeware software for laser range scanning
• Vision-Based Key Duplication
• input: a reasonably high-resolution photograph of a house key
• output: instructions for a milling machine to create a duplicate
• Robot Programming Contests
• Rat's Life
• IEEE MicroMouse

---

Most recently updated on Sept. 27, 2007 by Serge Belongie.