Fig. 4

Leaf tracking in a time-series of 3D maize plant segmentations. A Input data for the tracking algorithm, consisting of a time-series of 3D maize plant segmentations. Each segmentation includes ligulated leaves, which can be topologically ordered by increasing insertion height, and growing leaves. Ligulated and growing leaves are classified using stem height (red dotted line), which was smoothed over time. F Output data of the tracking algorithm, consisting of 3D + t reconstruction of the plant with time-coherent leaf ranks (numbers and colours), representing the order of appearance of the leaves. B, C, D, E illustrate the successive steps of the algorithm, which consists in assigning time-consistent ranks to segmented leaves. Each segmented leaf is represented by a rectangle, coloured according to its ground-truth rank (black = segmentation anomaly), and positioned according to its observation time and predicted rank. Only one third of the time steps are represented for visibility. B initialization of rank assignment by ordering the ligulated leaves topologically, C rank assignment after sequence alignment, D rank assignment after removing abnormal columns, E final rank assignment after adding growing leaves