Automatic inter-plant spacing sensing at early growth stages using a 3D vision sensor

An inter-plant spacing sensing system using a TOF (time of flight) of light based 3D sensor was developed. The 3D sensor was capable of capturing distance information, intensity and amplitude data in a single shot. The side view depth images captured were stitched together using distance information from a wheel encoder in conjunction with a feature-based image sequencing process for the stem location identification. One obvious advantage of the system over current color-based 2D systems was the use of depth images for plant identification, which was less sensitive to color variations. A covered cart was designed to prevent the sunlight from directly shedding on the plants and to reduce the interference from wind, which in turn made the system usable throughout the day. The vertical camera position was easily adjustable making the system suitable to work with plants at different growth stages.The use of side-view images made the system capable of detecting inclined plants and therefore, boosted the performance of the system in precisely locating the stem centers, which in turn minimized the measurement errors. The measurement accuracy demonstrated the system superiority over the current systems which make use of top-view images for inter-plant spacing sensing. The system achieved an overall mean root mean squared error (RMSE) of 0.017 m with a mean plant misidentification ratio of 2.2%. The coefficient of determination (R2) was 0.95 between the in-field manual distance measurements and the system distance estimates.

► Use of a state-of-the-art TOF (time of flight) of light based 3D sensor.
► Use of side-view images to detect plants inclined up to 15° from the vertical.
► Precise sensing of stem centers that minimizes inter-plant distance measurement error.
► Use of encoder data is pivotal in mosaicking process.
► Overall mean RMSE is 0.017 m and mean plant misidentification ratio is 2.2%.