A low-cost agriculture robot for dataset creation - software and hardware solutions

Abstract

The limited availability of public image datasets continues to be a significant obstacle for the efficient development and assessment of computer vision and machine learning algorithms in agricultural production. This paper describes a simple robotic vehicle whose goal is to collect datasets, i.e. images of plants in the early stages of development. The vehicle is designed for vertical photography of root vegetables (carrots, beets, parsley, etc.) that do not have a large leaf mass. The goal of the development is to make the production of vehicles cheap and to speed up the creation of freely accessible datasets by making such or similar vehicles. The robotic vehicle uses cheap and easily available components, and ArUco markers are used for the navigation itself, which enables simple and precise navigation.