Automatic leaf segmentation and overlapping leaf separation using stereo vision

Abstract

Farm management and crop quality assessment is becoming increasingly automated to keep up with demand. The physical examination of the plant leaves, stems and fruit can provide valuable information about a plant’s health. Automating the visual inspection through machine vision spawns challenges such as occlusions, irregular lightning and varying environmental conditions. In this paper, a plant leaf extraction algorithm utilising depth from a stereo vision sensor is presented. The algorithm tackles multiple leaf segmentation and overlapping leaf separation through synergising features such as colour, shape and depth. Depth is particularly used to measure discontinuities along its gradient in the disparity maps. The algorithm has a segmentation rate of 78% for individual plant leaves, over a range of complex backgrounds and changing plant canopies. The proposed algorithm was evaluated using 272 cotton and hibiscus plant images with results demonstrating that depth properties were effective in separating occluded and overlapping leaves, with a high separation rate of 84%. Leaf occlusion could be detected automatically without adding any artificial tags on the leaf boundaries. Furthermore, the results show a nearly identical performance for both types of plants (cotton and hibiscus) under various lighting and environmental conditions. The developed algorithm could be potentially applied to other types of plants that have similar structures to cotton and hibiscus.

Keywords

• Machine vision
• Stereo vision
• Depth segmentation
• Crop monitoring
• Leaves detection
• Overlapping leaves