Figure 2

Box diagram of module 1-3. General outline of the leaf angle processing toolbox: The structure of this toolbox comprises 3 basic modules (A-C), all of them controlled by individual graphical user interfaces (these GUIs are depicted in Figures 3, 4, and 5). Optional GUI-supported tools for HSV segmentation, stereo rig settings and result summary are not illustrated here. Module 3 comprises two alternative processing paths (left: via surface smoothing; right: via surface fitting). Alternative and optional modes in module 1 are indicated on the right sides as dotted boxes. Round boxes indicate the input and outcome of a process, rectangular boxes the processes themselves. Green boxes point to the subsequent processes in the other modules. The 3-d reconstruction (A) starts with the input of stereo images, calibration images and technical specifications. Outputs are rectified images and disparity maps on the one hand, which serve as the input data for the subsequent segmentation process. On the other hand, the 3-d point cloud data is transferred to the surface modeling process. With the data provided by the 3-d reconstruction the full or partial recognition of leaves is the intention of the image segmentation (B). On the base of segment-specific points clouds leaf surface structures are modeled in the third module (C) either using smoothing operations or bipolynomial surface functions. The resulting polygon mesh of the canopy provides a basis for further statistical analysis of particular plant traits like leaf angles or leaf area.