A two part question:
The output from DLC provides you with the x,y coordinates for any joint within each image.
However this is relative to the image size and also relative to the distance of the animal from the camera. Not a problem if all the video you have is the same size and distance to animal is standard, but it makes data analysis difficult if you don’t have either.
Is there a way of normalizing the x,y coordinates to deal with the difference in the image size (some are 640x360, some 1280x720 and others 1920x1090). I had initially thought:
xNorm = x/width of image
yNorm = y/height of image
But I wasn’t sure if that properly dealt with maintaining the aspect ratio of the image?
On the distance of the animal from the camera, once the image size is normalized and the x,y coordinates reflect a normalized position, is there a way of engineering each x,y position further to reflect the animals distance from the camera without knowing the focal length? Obviously I dont want to create a new x,y that destroys the actual physical size of the animal (e.g small/large). I was thinking of some type of ratio relative to the distance to 0,0 but my domain knowledge of image geometry has let me down.