Modelling and analysis of plant image data for crop growth monitoring in horticulture

Plants can be characterised by a range of attributes, and measuring these attributes accurately and reliably is a major challenge for the horticulture industry. The measurement of those plant characteristics that are most relevant to a grower has previously been tackled almost exclusively by a combination of manual measurement and visual inspection. The purpose of this work is to propose an automated image analysis approach in order to provide an objective measure of plant attributes to remove subjective factors from assessment and to reduce labour requirements in the glasshouse. This thesis describes a stereopsis approach for estimating plant height, since height information cannot be easily determined from a single image. The stereopsis algorithm proposed in this thesis is efficient in terms of the running time, and is more accurate when compared with other algorithms. The estimated geometry, together with colour information from the image, are then used to build a statistical plant surface model, which represents all the information from the visible spectrum. A self-organising map approach can be adopted to model plant surface attributes, but the model can be improved by using a probabilistic model such as a mixture model formulated in a Bayesian framework. Details of both methods are discussed in this thesis. A Kalman filter is developed to track the plant model over time, extending the model to the time dimension, which enables smoothing of the noisy measurements to produce a development trend for a crop. The outcome of this work could lead to a number of potentially important applications in horticulture.