Groups:Biosystems Engineering SDU:FroboMind vision integration thoughts

FroboMind Vision Architecture

For now this page is just a place to throw in random thoughts and ideas for how a vision architecture can be incoporated into the FroboMind framework (Morten's procrastination).

• Things which has to be investigated on the ROS side
  • How fast is the ROS IPC framework, ie. how small components can we divide our algorithms into using the ROS IPC, without causing to much overhead
  • When using the ROS IPC for moving images around, can we handle 4 channel images?
  • When using the ROS IPC for moving images around, can we handle more than 8bpp?
• Things which the framework shall be able to handle (Wish list)
  • Low-Level vision tasks
    • Segmentation(ExG, NDVI, Naive Bayes etc.)
    • Dynamic threshold
    • Morphological operations
    • Edge detection
  • Row following algorithms
    • Hough
    • Ransac
    • Custom
  • Feature extraction
    • How can we make a generic description of a feature??? (guess we need Thomas on this one)
  • Classifiers / Regression
    • Maybe we should make a ROS→ Weka interface (Thomas or Henrik, any ideas on if Weka is efficient enough for soft-realtime?)
  • Illustration
    • Plots
    • imshow
    • statistics
  • Simple OpenCV interface (I believe this is already very well integrated in ROS)
  • Artificial camera to allow images to be injected artificially
  • Matlab I/O for debugging (ie. input a variable from matlab to a specific ROS module and output the Result from the module to matlab)