Leaf Segmentation in Outdoor Environment Using A Low-Cost Infrared Camera

A camera system to monitor the leaf growth is a challenging topic due to illumination changes, shadows, and complex backgrounds. In such a system, a leaf segmentation is usually employed as an initial stage to extract the leaf from the backgrounds. This paper presents an implementation of a low-cost infrared camera system for leaf segmentation in an outdoor environment. A Raspberry Pi NoIR (No Infrared filter) with the infrared LED was employed to capture the Vetiveria zizanioides plant at a greenhouse. To handle the problems of illumination changes and complex backgrounds, the image was taken during the night time. This approach allowed the Otsu thresholding method to be used to effectively segment the leaf. The experimental results showed that the proposed method achieved a high segmentation performance, where the precision, recall, Jaccard index, and misclassification error were 0.9727, 0.9642, 0.9388, and 0.0065, respectively. The algorithm’s execution time implemented on the Raspberry Pi 3 Model B+ was 194 ms, which is suitable for real-time implementation. Furthermore, the increasing number of leaf pixels measured by the proposed algorithm over six days indicates that the method is reliable for leaf growth monitoring systems.