Articles
USING COMPUTER VISION FOR PUTTING FLOWER BULBS UPRIGHT
For the detection of the growing point of the bulbs the mechanical system will be provided with a PC-based computer vision system, equipped with a CCD-camera and a frame-grabber for image-acquisition.
Research on images of bulbs, manually put beneath the camera, shows that recognition of the growing-point of Freesia and Iris is quite feasible.
For other species, e.g. tulip, this is more difficult because of their more round shape (lit. 1).
For the detection of the growing point three different methods have been developed.
They are all in some way related to the characteristic shape of the outline of the silhouette of the bulb.
Preprocessing was done to remove roots and protruding bractlets from the silhouette to get a smoothed outline.
The first 2 methods are based on the a priori knowledge that in the image the growing point directs to the left or to the right.
This is due to the construction of the mechanical system which forces the bulbs to take the orientation that corresponds with their forward direction of movement on the conveyor shaft.
The first method computes the distance from the centrical pixel of the bulb to its far left and right end.
The second method counts the number of pixels in an area-fragment of equal horizontal length at both far ends of the bulb.
The fragment with the smallest number of pixels indicates the growing point.
The last method is independent of the actual orientation of the bulb.
The method computes the bend at each point of the outline.
The point that shows the strongest bend is considered to be the growing point.
Computation of the bend is accomplished by taking the straight distance between the end- points of a contour interval of fixed arc-length, which is moved on along the whole outline.
By changing the length of the interval the algorithm can easily be adapted to the typical differences in shape and size of the various species of flower bulbs.
The method showed to be very robust.