Tomato Development Monitoring in an Open Field, Using a Two-camera Acquisition System

Monitoring the growth of crops and estimating their yield allows the farmer to better manage the resources after the harvest. The existing methods for monitoring the growth of crops depend on remote sensing data or soil and weather information, which are prone to error due to adverse climatic conditions or insufficient information. Here we present an alternative approach, where the growth of the crop (tomato) is monitored from images captured in an open field. For this, the tomatoes are identified in the image using a segmentation procedure and their size is measured. This is a challenging task because of severe occlusions and poor contrast in the images. In order to increase the robustness of the segmentation procedure (based on active contours) and simplify the size estimation, we approximate the tomatoes as spheres in the 3D space, hence as ellipses in the image space. This model enables us to integrate a priori information about the shape of the object to be segmented and to avoid a complete reconstruction of the 3D scene. The automatic segmentation was evaluated by comparing the results with manual segmentation. For the cases with a reasonable amount of occlusion (less than 30%), good results were obtained, with an average relative error of 6.46% (expressed as a percentage of the tomato size). The metric reconstruction has also been evaluated. It was observed that the error on the estimated tomato radius was less than 5% for 91% of the cases. Finally, the complete system was tested. The size of the tomatoes was correctly estimated in 80% of the cases. In addition, preliminary studies have showed that the actual volume of tomatoes can be estimated from the calculated sphere radius, using a correction factor.