Two-stage approach for detecting slightly overlapping strawberries using HOG descriptor

Abstract

This article presents a new method for strawberry detection for use in a strawberry harvesting robot. The method is based on a histogram of oriented gradients (HOG) descriptor associated with a support vector machine (SVM) classifier. The detection involves two stages. First, strawberry-like regions are detected from HSV (hue, saturation, value) colour information. The HOG descriptor, calculated using five regions of interest (ROI), is input to an HOG/SVM classifier, which detects the strawberries. The performance of the model was verified by experiments. The vector sizes were effectively reduced and a higher detection speed was achieved without compromising accuracy (relative to conventional approaches). The proposed classifier achieves high detection accuracy (87%) in a reasonable run time, and can appropriately handle slightly overlapping strawberries.