Browsing by Subject "ToF-Kamera"

Now showing 1 - 1 of 1

Crop plant reconstruction and feature extraction based on 3-D vision
(2019) Vázquez Arellano, Manuel; Griepentrog, Hans
3-D imaging is increasingly affordable and offers new possibilities for a more efficient agricul-tural practice with the use of highly advances technological devices. Some reasons contrib-uting to this possibility include the continuous increase in computer processing power, the de-crease in cost and size of electronics, the increase in solid state illumination efficiency and the need for greater knowledge and care of the individual crops. The implementation of 3-D im-aging systems in agriculture is impeded by the economic justification of using expensive de-vices for producing relative low-cost seasonal products. However, this may no longer be true since low-cost 3-D sensors, such as the one used in this work, with advance technical capabili-ties are already available. The aim of this cumulative dissertation was to develop new methodologies to reconstruct the 3-D shape of agricultural environment in order to recognized and quantitatively describe struc-tures, in this case: maize plants, for agricultural applications such as plant breeding and preci-sion farming. To fulfil this aim a comprehensive review of the 3-D imaging systems in agricul-tural applications was done to select a sensor that was affordable and has not been fully inves-tigated in agricultural environments. A low-cost TOF sensor was selected to obtain 3-D data of maize plants and a new adaptive methodology was proposed for point cloud rigid registra-tion and stitching. The resulting maize 3-D point clouds were highly dense and generated in a cost-effective manner. The validation of the methodology showed that the plants were recon-structed with high accuracies and the qualitative analysis showed the visual variability of the plants depending on the 3-D perspective view. The generated point cloud was used to obtain information about the plant parameters (stem position and plant height) in order to quantita-tively describe the plant. The resulting plant stem positions were estimated with an average mean error and standard deviation of 27 mm and 14 mm, respectively. Additionally, meaning-ful information about the plant height profile was also provided, with an average overall mean error of 8.7 mm. Since the maize plants considered in this research were highly heterogeneous in height, some of them had folded leaves and were planted with standard deviations that emulate the real performance of a seeder; it can be said that the experimental maize setup was a difficult scenario. Therefore, a better performance, for both, plant stem position and height estimation could be expected for a maize field in better conditions. Finally, having a 3-D re-construction of the maize plants using a cost-effective sensor, mounted on a small electric-motor-driven robotic platform, means that the cost (either economic, energetic or time) of gen-erating every point in the point cloud is greatly reduced compared with previous researches.