3D monitoring for plant growth parameters in field with a single camera by multi-view approach

In recent decades, some photogrammetric methods for 3D monitoring of plant growth and structure parameters have been studied by the automated feature extraction and matching. In this study on growth analysis of sweet potato (Ipomoea batatas L.) plants at different fertilizer conditions, we proposed a convenient solution of 3D reconstruction by a single camera photography system based on Structure from Motion (SfM) method. Also, we handled effectively the noise problem by minimizing re-projection errors. The results of 3D models demonstrated that the average percentage error was a constant about 4.8% for plant height or decreased from 13% to 8% (leaf area index, LAI=3.5) for leaf number and from 19% to 12% (LAI=3.5) for leaf area with increasing in LAI, although each percentage error fluctuated, especially at low LAI. In contrast, the average percentage error in 2D image processing was 20% to 45% for leaf number and 60% to 90% for leaf area, and the leaf height was immeasurable. Comparing with the errors of 3D results, the errors of 2D images were much larger because 2D imaging had some problems, such as not being robust against occlusion of plant organs, and the ambiguity between object size and distance from the camera. On the other hand, we examined a method to calibrate the estimates from 3D models using the regression model between the measured value and the value estimated from 3D model. The regression models showed the linear and good estimation for leaf height (R-2=0.97 and RMSE=0.71 cm), leaf number (R-2=0.99 and RMSE=4.03) and leaf area (R-2=0.98 and RMSE=0.12 m(2)), in spite of use of data across a wide range in fertilizer supply and growth stage of sweet potato plants. The results demonstrated that 3D imaging technique in the study has the potential to remotely monitor plant growth status and estimate growth and structure parameters at various environmental factors outdoors.