Growth Measurement of a Community of Strawberries Using Three-Dimensional Sensor

Monitoring plant health is an important technique to guarantee high quality and quantity of agricultural production. In case of greenhouse-grown strawberry plants, farmers generally measure the plants’ height, number of leaves, and area of a single leaf to obtain growth information and estimate the plant’s health. These indices have been used successfully for many years in traditional strawberrygrowing. However, this method of monitoring plant health in an industrial-scale greenhouse would take up a lot of time. Furthermore, traditional growth information is not usually obtained every day. Even when the information is obtained, it involves taking only a sample value, which does not accurately report the condition of every plant. A circulating-type movable bench system for strawberry cultivation has been studied in Japan (Yoshida et al., 2008; Hayashi et al., 2011; Saito et al., 2012). With this cultivation system, all plants pass daily through a single location, called the access point, to be watered. The access point is thus the ideal place for continually and precisely obtaining information on the growth of strawberry plants. Machine vision is a promising technique for effectively monitoring every plant at the access point, since it enables non-destructive measurement. For effective plant health monitoring, many imaging techniques such as digital color imaging, multi-spectral imaging, thermal imaging and fluorescence imaging have been investigated (Takayama and Nishina, 2009). On the other hand, the availability of low-cost depth sensors that generate depth images is on the rise because of the development of natural user interfaces (NUIs), as seen in Microsoft’s Kinect sensor, which is commonly used as a motion controller for TV games. The accuracy of the Kinect sensor has been reported as a percentage of measurement error: between 2 cm to 2 cm is 90.9%, 82.9% and 81.2% for data along the x, y and z axes of the camera coordinates, respectively, using an uncalibrated Kinect sensor (Khoshelham and Elberink, 2012). In the camera coordinates, x and y axes are included in the CMOS plane of the camera. Namely, x axis is horizontal direction in the plane, and y axis is vertical direction. Direction of z axis is equivalent to the normal vector of the plane. For the leaf segmentation in rosebushes and the measurement of the leaf angle of ornamental plants, the low-cost depth sensor has been applied (Chéné et al., 2012). The Kinect sensor of these researches was equipped with an active-stereo system to obtain depth information. In 2014, new Kinect sensor was released which measures the depth by way of timeof-flight method. It is easily predicted that more researches will be conducted using the new sensor in the foreseeable future. We have developed a plant growth measurement method for strawberries using the active-stereo type Kinect sensor and have investigated the measurement accuracies of plant height and width and the area of leaves using a potted strawberry plant (Yamamoto et al., 2012). In this study, we propose an algorithm for a 3D measurement of a plant community of strawberries on a 1-meter long bench. We then report the results of a three-month observation of the plant community.