Towards an automated 3d reconstruction of plant architecture

Non-destructive and quantitative analysis and screening of plant phenotypes throughout plants' lifecycles is essential to enable greater efficiency in crop breeding and to optimize decision making in crop management. In this contribution we propose graph grammars within a sensor-based system approach to the automated 3D reconstruction and semantic annotation of plant architectures. The plant architectures in turn will serve for reliable plant phenotyping. More specifically, we propose to employ Relational Growth Grammars to derive semantically annotated 3D reconstruction hypotheses of plant architectures from 3D sensor data, i.e., laser range measurements. Furthermore, we suggest deriving optimal reconstruction hypotheses by embedding the graph grammar-based data interpretation within a sophisticated probabilistic optimization framework, namely a Reversible Jump Markov Chain Monte Carlo sampling. This paper presents the design of the overall system framework with the graph grammar-based data interpretation as the central component. Furthermore, we present first system improvements and experimental results achieved in the application domain of grapevine breeding.