Corn plant sensing using real-time stereo vision

Though some two-dimensional (2D) machine vision–based systems for early-growth-stage corn plant sensing exist, some of their shortcomings are difficult to overcome. The greatest challenge comes from separating individual corn plants with overlapped plant canopies. With 2D machine vision, variation in outdoor lighting conditions and weeds in the background also pose difficulties in corn plant identification. Adding the depth dimension has the potential to improve the performance of such a sensing system. A new corn plant sensing system using a real-time stereo vision system was investigated in this research. Top-view depth images of corn plant canopy were acquired. By processing the depth images, the algorithm effectively updated the plant skeleton structures and finally recognized individual corn plants and detected their center positions. The stereo vision system was tested over corn plants of V2–V3 growth stages in both laboratory and field conditions. Experimental results showed that the stereo vision system was capable of detecting both separated and overlapped corn plants. During the field test, 96.7p of the corn plants were correctly detected, and plant center positions were estimated with maximum distance errors of 5 and 1 cm for 74.6p and 62.3p of detections, respectively. © 2009 Wiley Periodicals, Inc.