A preliminary field evaluation of an automated vision-based plant geometry measurement system

[Introduction]: Machine vision is commonly reported for use in automated plant-based applications such as growth monitoring, stress detection, species identification and fruit harvesting. Geometric measurement of living plants using machine vision commonly requires depth perception, which typically entails establishing correspondence between multiple images. The complexity of such a task in an unstructured environment, such as foliage, often restricts automation of the process. Automated measurement of the geometry of young plants has been reported using laser range finding and image processing (Kaminuma et al. 2004) or multiple camera views (Lin et al. 2001). Noordam et al. (2005) compared automated methods of locating the main stem of a rose plant using methods including stereo imaging, structured lighting and X-ray imaging, while Ivanov et al. (1995) reported using stereovision and a manually-operated 2D digitiser to model a maize canopy. However an automated vision system for measuring geometry of complex plants in the field is yet to be reported. Plant geometry is a significant factor for irrigation purposes in cotton because the distance between main stem nodes indicates water stress in a growing cotton plant. However, manual measurement of internode length is a tedious task for even a small number of plants. An automated method for large scale measurement of plant geometry in the field would provide information about crop irrigation requirement as well as spatial and temporal variability in crop stress and growth. This research aims to develop a real-time and automatic machine vision sensor for measuring structural parameters such as internode length for plants in a growing cotton crop