Adjust module detection preprocessing

[fallscameron01]

Description of Problem

Circle detection is finicky and requires good preprocessing. Module detection pre-processing needs to be adjusted to ensure background is not included in circle detection and holes on module can be reliable detected.

Requirements

• [x] Adjust existing preprocessing to improve accuracy of circle detection.
• [x] Implement new preprocessing if necessary. (examples: radial blur, color filtering, increasing brightness)
• [x] Filter out circles based on other criteria. (examples: depth, distance from text, distance from edge of image)

[fallscameron01]

Potential ideas:

• increase brightness in areas where depth is small, so brightness would be increased on the mast
• increase contrast
  • should not decrease darkness however, since this could make circles undetectable
• unsharp masking
  • which actually increases sharpness https://en.wikipedia.org/wiki/Unsharp_masking#Implementation
  • can potentially be done with a simple kernel
  • maybe do only in areas where depth is small? Combined with blur would mean mast is sharp and background is blurred.
  • may harm circle detection, needs to be tested.
• color filtering based on the blue on the mast
  • potentially difficult since sky is similar blue. would need to be nearly perfect and also consider any contrast/brightness increases
• run depth image values through np.histogram to create groups of depth values
  • sample depth values so creating histogram would not take a long time. example: depth[::5, ::5] would take every 5th value
  • filter out circles that are far away
• filter circles based on location in image
  • use existing text detection to determine rough placement of mast, filter circles based on proximity to text
    • may not work in every situation. would only work when text is in the image which may not be the case if we are close to the module.
  • filter out circles near edge of image
    • would need to be more directly looking at module to detect circles

[fallscameron01]

Bilateral filtering: https://docs.opencv.org/master/d4/d86/group__imgproc__filter.html#ga9d7064d478c95d60003cf839430737ed