cameraLoc
environment
qt5-mingw32, opencv3.0
function
recognize a QR code locator in an image, then tranport the center into world point, to locate the object the locators attach to
feature
can deal with some perspective distortion and rotation, but too much is not guaranteed.
menthod
- the QR code locator is designed cleverly. Even when the locator is rotated, we can still get a horizental line bar and a vertical line bar, and each line in these two bars have the same ratio: 1black-1white-3black-1white-1black. center lines of two bar intersect at the locator center
- when I test this property in practice, I find it hard to have a continous bar, or there exists other lines with correct ratio outside the bar. I write a filter for one dimention situation to solve this.
- also, I find it hard to have exact 1:1:3:1:1. I believe once the ratio error comparing to all found length(note to get more possible lines I compare all found) exceed tolerance, the proportion is wrong. Above codes can be found in Detector::centerRecognize
- also, using opencv adaptiveThreshold is another point. See Detector::getImagePoint. I have also tried other ways like gussian blur, erode, but these won't give better results.
- At last, I find my method works not so well, so I read zxing source code, which is an open source project for QR code recognization. I extract the locator recognization part, and convert them to qt(c++)...Then it works like a magic. See finderPattern and finderPatternFinder class for recognization details. using it is quite simple: FinderPatternFinder finder(src); finder.find(); see Detector::centerRecognize for details about how I use it.
