Sample images. obstacles on the test field, cart paths on a golf course, street scenarios:
using a web cam
using the PGR Butterfly stereo camera
Run-time vision. Add additional compute-nodes to ERTS; prototype
image "feeds" and telemetry.
Matlab models
Horizon Detector. This is the simplest model, needed to test Matlab/SyncFS integration
Vanishing Point
Dominant Features as a means of obstacle detection
Other models from vision literature
Stereo Ranging: studies of path finding based on simple ranging
CMUcam/CartFS integration. The CMUcam and its tracking software
were originally developed using a client/server model. This should be
ported to the CartFS model, with a component to handle the camera's
serial connection.
CartFS Frameworks for Matlab, Java (Dimitar, Fan), C/C++,
Scheme (Johnson, Kuper), and any
other language of interest.
More to come
ERTS function. Build ERTS capabilities and configurability
GPS following. Existing GPS followers are relatively primitive;
improvements are needed for turning maneuvers, intermediate range path
planning, etc.
Braking. Braking was added to ERTS's control function last summer.
Low and intermediate brake control remains to be developed
ERTS vision. A major goal in ERTS development this Fall is adding
vision to its sensory system. The individual projects listed below are
hightly iterrelated. As the project descriptions are expanded, look for
cross links.
Point Grey Research (PGR) Bumblebee (BB) Integration.
The Bumblebee is a stereo camera that will likely be a main source of
vision imagery.
Windows XP API (Rich Lewis).
Use PGR libraries to
capture, record, and play back image pairs;
generate stereo disparity maps
work with contacts in the IU Advanced Visualization Lab (AVL) lab to
play back recorded stereoscopic imagery using a 3D visualization
system; prototype an architecture for real-time stereo telemetry
Linux API (Dave Bender).
Bring Lucas Johnson's Linux/IEEE 1394 API on line.
Capture, and record image pairs; generate stereo disparity maps;
develop a CartFS proxy component for communication with a navigation
system
Vision networking. Prototype ad hoc networks for existing cameras.
Propose a uniform design models for sensor-fusion networks.
I am looking for groups of two or three students to commit to the following
functional objectives, starting this Spring. These would be continued under
indpendent study or as thesis projects.
Propose your own experiment.
Plenoptic Sensor.
Use plenoptic imagery for obstacle tracking.
In collaboration with Andrew Lumsdaine.
See examples [HTM]
Drive-by-Voice.
Develop a core system for voice interaction on ERTS.
Probably in collaboration with Matthias Scheutz's lab.
Insect vision.
(Tentative) Deploy an experiment based on the work of IU Neuroscientist
Robert Taimond de Tuyter van Steveninck.
Note: I haven't contacted Prof. de Tuyter van Stevenicnk about this yet, but I have
chatted with him briefly and he expressed interest.