MIME-Version: 1.0
Server: CERN/3.0
Date: Monday, 16-Dec-96 23:20:57 GMT
Content-Type: text/html
Content-Length: 3313
Last-Modified: Saturday, 09-Mar-96 21:37:39 GMT
Suggested Projects for CS664
Suggested Projects for CS664
Robotics
- Edge Clustering Develop algorithm for clustering of
edges that represent a single object.
- Performance Simulations Investigate how high-performance vision
and image processing applications perform on modern architectures, vision
applications tend to use large data structures (images) that can exhibit both
good and bad spatial locality. Ideally, one would have a processor simulator
that would be configureable: cache size, cache associativity, memory size,
processor speed, instruction set, etc. The simulator should be able to
generate statistics such as cache hit/miss ratio. Hopefully the simulator
could be extended to simulate a Shared Memory Multiprocessor (SMMP). Starting
from scratch will be very hard, it may be better to adapt an existing program
to our needs. Starting points: UW.
- Image Databases There is currently a project here at Cornell to
develop a general purpose image database that will allow "intelligent"
searching for images based on a large number of image features. We need
people to implement different measures, for basics such as texture pattern
matching to face recognition. Starting points: the competitors.
- Camera Gaze Control Create or adapt an existing object tracking
algorithm so that the output can be used to control the pan/tilt positions of
a desktop camera. The robotics lab is developing a relatively-cheap pan/tilt
hardware for a standard CCD camera. We would like the user to be able to
click on a portion of the image and have the camera pan/tilt to keep the
object centered in focus.
- Find a Cool Use for Stereo Vision Existing programs allow us to
generate good depth maps at near-real time speeds (5-15 FPS). We would like
someone to create a "cute" application using the robots or some other platform
that would use the depth maps to perform some task, such as navigation or
collision avoidance (these are not actually the same, currently the easiest
way for the robot to navigate is for it to go until it "crashes" into
something, back up a little, then turn).
- Stereo Camera Calibration Develop a robust, reliable procedure
for calibrating a pair of stereo cameras, using a feedback loop between the
output of the cameras as they view a test pattern and their next adjusted
view.
- Text Segmentation Develop a system that can detect text in an
image or series of images. This is not the same as OCR, where one tries to
take an image of printed text and recover the text; rather, you should
concentrate on how to find text in an image. Typical examples of texts that
one might want to identify from a video signal include the bylines on network
news and scores and player stats from sports games.
- Human Face Segmentation Basically, find the objects in a scene
that look like human faces. How many people are there? How certain is this
answer? Starting point:
Univ. Maryland, also work at CMU by Kanade's group.