Algorithms to estimation of size and shape tomato using Artificial Vision Techniques

Abstract

This paper describes the development of estimation algorithms for the size and shape of tomato fruits, implemented on a portable electronic system used in greenhouses, with the purpose to classify and collect of tomatoes milano and chonto type. The algorithms were implemented in a Raspberry-Pi embedded card through the use of free software libraries (Python, OpenCV, Pillow and Scikit-Image) and using a Pi-Camera for real-time processing, taking into account the parameters of classification defined in national (NTC1103-1) and international (USDA) standards. Were applied classification algorithms based on Canny to estimate the size of the tomato fruit, as well as techniques based on estimation of the eccentricity and statistical moments descriptors for the measurement of the shape, which were evaluated with a performance superior to 90 %. Classification techniques of tomato shape and size using artificial vision overcame subjective techniques of visual and tactile type classification carried out by farmers and specialized technicians.