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
    1. Edge Clustering Develop algorithm for clustering of edges that represent a single object.
    2. 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.
    3. 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.
    4. 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.
    5. 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).
    6. 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.
    7. 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.
    8. 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.