The problem of quantifying the growth dynamics of plants from time-lapse images is important for determining variety-specific characteristics and subsequent breeding. The goal of this paper is to address this problem and reconstruct three-dimensional images such that growth becomes visible. To this end, data of growing plants were collected in the form of multi-angle time-lapse images. In each image separately, the plants' outline was detected with a GVF-Snakes algorithm. The success of this algorithm is ensured by pre-processing steps and placing the initial contour based on color specifics of the image.
The two-dimensional contours are then combined to reconstruct a three-dimensional contour in the camera reference frame. To obtain this reconstruction, points along the contour are matched using a RANSAC-based algorithm to estimate the essential matrix. The three-dimensional reconstructions show information about size, height, leaf structure, and growth of the plant, but the stability of the method has to be improved. Finally, an attempt is made to combine the GVF-Snakes with the point-matching algorithm, however, the convexity of the coplanarity constraint should be explored further.