Abstract
When dealing with unmanned agricultural vehicles (remotely-controlled vehicles, robots), vision systems are a key-factor for implementing field-solutions having direct interactions with crops.
Among all the possible information given by a vision system, the punctual estimation of the canopy volume is surely an interesting parameter: it is related to the crop vegetative status and, hence, it is fundamental for performing and setting-up properly some important field-operations (e.g., pruning/thinning, spraying). A system able to recognize the canopy volume can provide either the input-signals for implementing a robotic real-time site-specific farming system or relevant information for a proper crop management. However, there are many practical difficulties in the field implementation of such a system: complex canopy shapes, different colours, textures and illumination conditions with projected shadows.
Terrestrial/aerial vision systems working on visible-light wavelengths and/or 2D-images of crops, although capable of excellent performances, have a computationally-heavy post-processing; therefore, they are unsuitable for implementing low-cost real-time servo-actuated cropping systems (e.g., robotised sprayers). Instead, a vision system composed by two LiDAR sensors aligned vertically, scanning the same targets, could give a sort of stereoscopic vision, here named ‘‘lateral-linear-stereoscopic vision”.
The aim of this study is assessing the opportunity to use such a system on an automatic or autonomous/robotised implement by performing some preliminary tests in a controlled environment. The resulting system is independent of the lighting conditions (it works also in the dark), is highly reliable (no projected shadows) and data processing is very fast. Although further studies are required to overcome the issues that could arise in a future field implementation, this system has all the premises to be successfully embedded in an automatized monitoring system.