Towards 4D Crop Analysis in Precision Agriculture: Estimating Plant Height and Crown Radius over Time via Expectation-Maximization

 

L. Carlone, J. Dong, S. Fenu, G. C. Rains, F. Dellaert, "Towards 4D Crop Analysis in Precision Agriculture: Estimating Plant Height and Crown Size over Time via Expectation-Maximization", ICRA, Workshop on Robotics in Agriculture, 2014.

 

mole2D

 

Abstract- In previous work [1] we showed how to apply modern sensor fusion and computer vision techniques to obtain a dense 3D reconstruction of a crop for precision agriculture. In this work we consider the case in which we have multiple 3D reconstructions, obtained from sensor data collected over several weeks. From this collection of dense reconstructions, we want to estimate how the size and height of each plant evolve over time. The problem is challenging since the 3D reconstructions may contain very partial views of each plant. Moreover, the presence of multiple plants (and background) requires solving the data association problem, which makes our goal even more challenging. We propose a general probabilistic model to estimate shape and appearance of objects (the plants) using factor graphs. Then, we tailor the formulation to precision agriculture, and we show that the choice of a suitable parametrization and the use of expectation-maximization enables fast inference on plant growth. Our approach provides high-level description of the status of each plant, and this can inform the farm manager and enhance awareness. Current results are extremely encouraging, and open several avenues for future research. 

 

 

[1] J. Dong, L. Carlone, G. C. Rains, T. Coolong, and F. Dellaert, “4D mapping of fields using autonomous ground and aerial vehicles,” in 2014 ASABE and CSBE/SCGAB Annual International Meeting, 2014. 

 

 

Supplementary material

  • Extra results and visualizations