Automatic Detection of the Main Vine and Branches of Tomato Plants Grown in Greenhouses

Abstract

An algorithm for real-time detection of plants main vine and branches using the Distance Regularized Level Set Evolution (DRLSE) algorithm is presented in this paper. The algorithm tracks the target plant in the foreground from a set of seeding points automatically located on the main vine. The location of the seeding pints is determined in two different ways using the Hough transform and template searching. Test images are captured inside a greenhouse and include a huge amount of clutter from adjacent plants. The clutter creates a complex environment for traditional segmentation techniques to detect branches of the plant in the foreground. Images are normalized to speed the execution time of the DRLSE and protect the algorithm from local minima traps. Results demonstrate the capability of the algorithm in detecting the main vine and major branches. This algorithm is good for real-time maintenance (i.e., deleafing) using robots inside greenhouses.