Research on estimation of plant height on vegetables by RGB-D camera for Smart Agriculture

<p>Plant height, or the natural height from the ground surface to the top of a plant, is an essential metric for estimating the growth status of cultivated vegetables. The variation of plant height over time indicates the intensity of plant vigor, determining the amount of irrigation and fertilizer application during cultivation. Recently, various 3D-shape measuring instruments have been used for automatic plant height measurement, among which the RGB-D camera is reported to be relatively inexpensive and easy to acquire depth data. The previous study proposed an RGB-D camera-based plant height estimation method that assesses the plant height by averaging the 5% and 10% uppermost pixels. However, some unexpected inclusion into the pixels with the leaves closing to the top of stems lowers the estimated height, resulting in low accuracy (84%~100% and 45%~100%, respectively). This paper presents a novel algorithm to search the tops of stems to estimate the plant height. An RGB-D camera was placed directly above the target plants to continuously take pictures from the planting stage when the density of leaves was relatively small. The algorithm marked the highest pixel as the first top of plant stems and excluded a specific range around it before continuing the search, aiming to eliminate the influence from the lower leaves. This process was repeated until the number of found tops agreed with the number of stems; the plant height was then calculated by averaging the height of stem tops. As an evaluation experiment, the estimated height was compared with the measured value, showing a high accuracy of 94%~100% in our method.</p>