Computer Vision

Home     Syllabus

Lectures     Examples     Links

Some of the uses for computer vision include:

• Using stereo and sequence images for uncertain 3-D environment modeling.
• Using vision for robot path determination.
• Identifying objects or people in images.
• Monitoring areas for intruders.
• Understanding image contents.

Interesting problems to solve:

• Acquiring Images
  

  The first problem is getting images into the computer and manipulating them via the computer software. This requires that the operating system has the correct drivers, and that the libraries and API documentation is available for controlling those interfaces via the computer vision code.

  
  
  
  
• Finding Features in Images
  

  There are many types of features and feature finding algorithms in literature, but some are worse than others for the complexity of their algorithms, their alogorithm execution time, and the usefulness of the resulting features and feature properties for achieving the vision problem goals.

  
  
  
• Feature Matching Between Images
  

  There are many ways to match features, generally depending on the properties of the given features, but quality of match, accuracy, and compute time required are significant problems in doing those matches.

  
  
  
  
  
  
  
• Stereo Views
  Features in stereo view image pairs must be matched to identify changes in offsets to the feature location in the images and the offsets can then be used to estimate the relative distance of the source of the image features from the cameras.
  
  
  
  

  Left Camera View -------	------- Right Camera View
  Notes: Note the differences in camera image quality to be overcome in matching between images, including color bias as above. Also note the "nose" (stick between cameras) which helps in the automation of camera calibration for angles to 3-D distances, but only if it can be matched (registered) in the two images. The overall "clutter" is one of the more difficult aspects for computer recognition of real scene objects.
  Left Edges View -------	------- Right Edges View
  Notes: Note the inconsistency in "edgeness" in each picture and between pictures. Also note the "nose" (stick between cameras) Whose edges don't really make it easier to match the ends between images. The overall "clutter" is still dominating in the "edgeness" versions.

  
  
  
• Successive Views
  

  Features in successive views need paired to find changes in location and estimate movement of the target objects and the camera position.
  
  

  Full color	Greyscale	"Edges"
  Notes: Outdoor Scene Sequence (real clutter, pixel color vectors) Greyscale (single pixel shade integer) "Smoothness" vs "Edgeness" values

  
  
  

• Related Subject Areas
  
  
  
  • 2-D and 3-D model representation forms and manipulation techniques
  • Real error tracking and control in awkward and large-scale numeric techniques
  • Generalized characterization of the similarity of objects with nonhomogeneous attributes
  • Extreme performance hardware architectures (multi-core, GPUs, and FPGAs) for AI, vision, speech understanding, and other specialized algorithms.
  • Asynchronous algorithm design and development techniques.
  • Path finding in 3-D models
  • Actor/Event models for tracking dynamic obstructions
  • Cerebral cortex and mental modeling for matching complex patterns
  • Concept nets for "understanding" input patterns
  
  

Other Links:

Computer Science Department Homepage
Information Technology Department Homepage
New Mexico Tech Homepage