[en] The aim of this study was twofold. The first goal was to acquire high accuracy stereoscopic
images of small-scale field scenes, the second to examine the potential of plant height as a
discriminant factor between crop and weed, within carrot rows. Emphasis was put on how to
determine actual plant height taking into account the variable distance from camera to
ground and ground irregularities for in-field measurements. Multispectral stereoscopic
images were taken over a period of 19 days starting one week after crop emergence and
seven weed species were considered. Images were acquired with a mobile vision system
consisting in a filter wheel based multispectral camera and a video projector. The
stereoscopy technique used belonged to the coded structured light family.
The stereoscopic acquisition method yielded good results despite the numerous
stereoscopic difficulties exhibited by the scenes. A plant height parameter as opposed to
distance from camera to plant pixels gave better results for classification (classification
accuracy of up