Sweet Pepper Detection

Visual Detection of Occluded Crop: for automated harvesting

Authors: C. McCool*, I. Sa*, F. Dayoub, C. Lehnert, T. Perez, and B. Upcroft

This paper presents a novel crop detection system applied to the challenging task of field sweet pepper (capsicum) detection. The field-grown sweet pepper crop presents several challenges for robotic systems such as the high degree of occlusion and the fact that the crop can have a similar colour to the background (green on green). We present a novel system which uses a conditional random field (CRF) and a set of traditional computer vision features (such as local binary pattern and HSV) to perform crop segmentation. This algorithm is able to detect 69.2% of sweet peppers in data captured from real in-field sweet pepper farms which is an impressive result given that the average detection accuracy of people viewing the same colour imagery is 66.8%.

A version of this paper can be found here.

Data set

To encourage other researchers to explore this problem we have made available the data set used for this paper. If you would like access to this data please send an email to the following address with the title "Re: Sweet Pepper Dataset".