Apple Detection for Harvesting Robot Using Computer Vision

Author Name(s): Abhipray Paturkar, Gourab Sen Gupta, Donald Bailey
Author Email: A.Paturkar@massey.ac.nz

Abstract

Development of fruit detection has always been a challenge for a harvesting robot. Detection of fruit in an unregulated environment with change in illuminations affects the harvesting robot’s efficiency. This paper presents the improvement of robotic apple harvesting by enhancing apple detection and visibility by handling occlusion in outdoor conditions. For detection of the fruit, we developed an image processing method that is robust to change in illumination and partial occlusion of the apples; we used a stereo camera pair for image acquisition. To detect the fruits efficiently, global thresholding, based on chromaticity approach, is used. The results demonstrated that fruit detection achieved accuracy of 95% in sunny condition and handled partial occlusion of the fruit. This study will be helpful for detecting apples for harvesting robots.

Keywords Computer Vision, Stereo Vision, Harvesting Robot, Apple Detection, Occlusion Handling

Introduction

Advancement in precision agriculture has improved the quality of the products, productivity and yield of the field and at the same time reduced production, and labor costs [1]. Regardless of these advancements, many of the agricultural tasks are still being done manually, which are tedious, difficult and monotonous operations but are still complicated for robots to do. An important part in successful harvesting robots is their capability to process the sensory data, specifically, their ability to evaluate and understand video input. Practically, combination of video input and smart computer vision algorithms might help to perform many other operations and improve harvesting robots considerably. Nonetheless, the issues related to computer vision in outdoor conditions are numerous such as immensely unstructured conditions with large degree of ambiguity; fruits with different shapes, colours, textures, reflectance properties and sizes and serious occlusions, are the part of the issues that computer vision systems might face. These computer vision systems are yet to reach full potential to implement solutions at practical level [1].

 

Conclusion

In this paper, the improvement of apple detection for harvesting robots was presented and detailed experiments were performed. The vision system consists of a stereo camera and a PC. There are two main contributions; an apple detection algorithm that can deal with change in lighting condition and can handle partial occlusion. The chromaticity approach with red coefficient was effective for segmentation and detection. Total 90 images, captured under varying lighting conditions, were tested. 85% to 95% of the apples were correctly detected under different lighting conditions. Partial occlusion of apples was handled. The developed apple detection algorithm could be used in harvesting robots. This approach provides an effective solution to the detection of apples and could be used for other fruits or objects.

1,136 total views, no views today

Download PDF File

About the author: admin