Image Processing and Analysis of Leaf Movement in Mimosa Pudica

Abstract

This thesis has the aim of develop and improve methods used in an ongoing research project at the University of Stavanger, where the plant mimosa pudi- cas response to change in its illuminative conditions is being examined. The intention is to use these methods to analyse the mimosa pudicas responses, as well as exploring the possibility of relating them to a control theoretic viewpoint. An image processing routine, using the HSV color model and triangle intensity threshold segmentation, was developed to segment time-lapse image series of the Mimosa Pudica - quantifying the plants image pixel count as a metric. The result was used to estimate the ARX, ARMAX and Box Jenkins black box models, with the plants illuminative condition and image pixel count being used as the systems input and output. A routine for a second metric was also developed, using the method of Farneback to estimate dense optical ow, describing the plants movement orientation in the image plane. The automatic non-parametric image segmentation routine performance was evaluated on 7 test images, obtaining a DICE similarity coe cient of 0:932 0:0175. Several ARX, ARMAX and Box Jenkins models was estimated without being able to capture the systems dynamics. The routine describing mimosa pudicas movement orientation was during the experiment found to give a seemingly correct description of a small time-lapse sequence. The successfully development of a automatic non-parametric image segmentation routine will remove the need of manual intervention. The observed dynamics in the mimosa pudica indicates that the changes in the illuminative condition could be working as a disturbance on the plant. The dense optical

ow experiments did also shown patterns of the mimosa pudica starting its response to the shift in the illuminative condition, before the actual shift occurred. This suggests that a complex structure could lie behind the plants response to the illumination regimes.