MACHINE VISION FOR THREE-DIMENSIONAL MODELLING OF GERBERA JAMESONII FOR AUTOMATED ROBOTIC HARVESTING

An image processing algorithm was developed that identifies flower stems from Gerbera jamesonii in images. Flower stem regions in the image were split into several independent parts at characteristic positions. Parts which contained crossings and noise where eliminated and the remaining parts were put together by a collinearity criterion to form a flower stem object. The position of the identified flower stem object was computed by means of triangulation using information from stereo images. The result was a three-dimensional object of the flower stem which should act as input for an automated robotic harvesting process. Results show that in 72 % of the images all flower stems were identified correctly. Looking at the stereo cameras in 61 % of the image pairs all flower stems were identified correctly in both images. Looking at all camera positions the accuracy increase to 97 %. Therefore it can be concluded, that the algorithm is accurate enough to enable automatic robotic harvesting of Gerbera jamesonii.