Skip to content
This repository has been archived by the owner on Dec 31, 2022. It is now read-only.

Generic Detector

Jump to bottom
Alex Carter edited this page Sep 23, 2018 · 4 revisions

STATUS: 90% Competition Ready
LAST UPDATED: 1/1/2018 w/ DogeCV 1.1

LEGACY DETECTOR. IS NOT A DogeCVDetector AS OF YET

About

This detector is made for those who want to detector another object this year (such as relic) or want to get into OpenCV. This class can be thought of as a PixyCam+ as it has the same limitations but has a few more options to increase reliability. This is built off of the Jewel Detector.

Details

This detector uses color filters to generate a mask. Using these masks we find contours (edges) and draw boxes around them. We then go into scoring these boxes.

To score these boxes, we calculate the difference from "Perfect" in different attributes such as ration (height/width of the box) and area of it. That way the box that is closest to each of these perfect values is deemed the best result.

The two attributes we use for this scoring are ratio of the cube and area of it in pixels. We uses these two attributes to find a perfect square that is the size of a jewel (or the largest object). Both of these values and how important are they to the final score can be tuned.

Usage

This detector is made to be a introduction to making your own detectors or filling gaps in this year's game.

Tuning

Still have to do this. Theres alot :/

Parameters

  • colorFilter - Color filter to use (See Color Filter API Wiki Page).
  • downScaleFactor - representing how much to downscale each frame.
  • detectionMode - Mode used to score results,
    • MAX_AREA returns the largest objects in the frame (Good for quick and easy since it doesn't need tuning, but less accurate)
    • PERFECT_AREA returns the objects closest to a desired area (Needs tuning, use debugContours to get areas)
  • rotateMat - Rotate the image when processing (wont be visible on preview, change this if you see detections working horizontally) [Usually: Landscape = false, Portrait = true]
  • areaWeight - How much does area affect the results (should always be a decimal under 0.1 due to the large numbers returned by areas)
  • perfectArea - The area of an object (in pixels) that is deemed "perfect" for jewels.
  • debugContours - Show debug information for contours in the image, including areas.
  • maxDiffrence - Max diffrence for objects (lower is more restrictive, recommend 10-20)
  • ratioWeight - How is the ratio of the object's width and height weighted (recommend 10-20)
  • minArea - Min area for objets (this depends on downScale, but I recommend 500+);

Returned Data

Currently this detector returns the following:

  • getLocation() - Get center location of the chosen object (Point)
  • getRect() - Get the full rectangle chosen

Changelogs

Clone this wiki locally