Robot Vision System

1
Robot Vision System

• Major Phases in Robot Vision Systems
• A. Data (image) acquisition
• Illumination, i.e. lighting consideration
• Lenses, and Cameras
• Digitizers
• B. Pre-processing
• Enhancement, i.e. smoothing, edge detection
• Segmentation (Binarization)
• C. Recognition
• Feature extraction
• Pattern matching
• D. Part Manipulation
• Choice of robot, gripper
• Orientations of gripper, camera, part w.r.t robot
  base

2
Vision System Pre-Processing

• Two Approaches
• Frequency Domain - refers to the collection of
  pixels (complex) resulting from taking the
  Fourier transform of an image. (not covered in
  this course)
• Spatial Domain - refers to the collection of
  pixels that forms the image. The methods are
  procedures that operate directly on these pixels.
  (this is the approach used in this course)
• Mathematically,
• g(x,y) h f(x,y)
• where f(x,y) is the input image
• g(x,y) is the resulting image and
• h is an operator on defined over some
  neighbourhood at
• pixel (x,y).

3
Vision System Pre-Processing

• Spatial Domain Approach -
• g(x,y) h f(x,y)
• Usually, h is a square array of numbers called
  convolution mask, or templates, or windows or
  filters.
• Example h is a 3x3 mask
• g(x,y) w1 f(x-1,y-1) w2 f(x,y-1) w3
  f(x1,y-1)
• w4 f(x-1,y) w5 f(x,y) w7
  f(x1,y)
• w7 f(x-1,y1) w8 f(x,y1) w9
  f(x1,y1)
• Example given in class

4
Vision System Pre-Processing

• Smoothing Reduce noise and other effects that
  are present in the image which are results of
  sampling, quantization, transmission, etc.
• Methods
• Neighbourhood averaging
• Given an image f, the resultant smoothed image g
  is obtained by averaging the intensity values of
  a neighbourhood centered at pixel (x,y), ie.
• Where S is the neighbourhood set defined at
  pixel (x,y)
• p is the no. of pixels in the
  neighbourhood.
• Example given in class

5
Vision System Pre-Processing - smoothing

• Median Filtering
• Recall the median (m)of a set of values is such
  that half of the values in the set are less than
  (and equal) m and half the values are greater
  than m.
• e.g. Set (10, 20, 20, 20, 100, 20, 20, 25, 15)
• sorted gt (10, 15, 20, 20, 20, 20, 20, 25, 100)
• Idea is to force pixels with very distinct
  intensities to be more like their neighbours.

6
Vision System Pre-Processing - smoothing

• Example
• Original
• By averaging
• Median Filter

7
Vision System Pre-Processing - Edge Detection

• Edge indicates a change in intensities within a
  window
• Edge detection techniques involve derivative
  operators
• The gradient at pixel (x,y) in an image is
  defined as a 2-D vector,
• The magnitude of the gradient
• The direction of the gradient
• For edge detection, we are interested in the
  magnitude only.

8
Vision System Pre-Processing - Edge Detection

• Common First order Gradient Operators
• First order difference between adjacent pixels
• Masks

9
Vision System Pre-Processing - Edge Detection

• Common First order Gradient Operators
• Sobel Operators
• Image window centered at (x,y)
• Mask_x Mask_y
• Gradient magnitude

10
Vision System Pre-Processing - Edge Detection

• Laplacian Operator
• A second order derivative operator,
• For digital images,
• Image window centered at (x,y)
• The Laplacian Mask
• The gradient magnitude
• Note Laplacian operator is extremely sensitive
  to noise

11
Vision System Pre-Processing - Thresholding

• Problem Given a single value of intensity (L)
  such that the resultant binary image B, is given
  by
• where f(x,y) is the input grey level image.
• Question is to determine L?

12
Vision System Pre-Processing - Thresholding

• Algorithm
• Step 1 Obtain the intensity histogram of the
  image
• Step 2 Find the valley of the histogram. The
  grey level value corresponding to the valley is
  the value of L
• Step 3 Assign the value of zero or one to each
  pixel