Fig. 3

Berry time-lapse tracking pipeline. A 10 segmented RGB images sampled from a 68 images time series representing the evolution of one bunch over time. Raw images were captured with a median interval of 8 h. B Scatter plots of the coordinates of the berry ellipses centres detected at two time steps (\({t}_{15}\); blue circles, \({t}_{46}\); red crosses), before (left) and after (right) registration. The distance metrics \(D\) between the two point sets is given below each plot. C Heat map of the distance matrix, storing the distance between all pairs of time-points after registration. Red points correspond to matrix values below the threshold \(\theta =8px\). D Matching tree, determining the order in which labels are propagated during tracking. Each rectangle represents a time-step. The highest one corresponds to \({t}_{root}\), used to initialise tracking labels. E Labelled segmented images after tracking. Each colour corresponds to one tracking label. Segmented berries without label (no match found with \({t}_{root}\)) are drawn as red empty ellipses