Plant architectural traits are important factors in determining grain and biomass productivity of sorghum (Sorghum bicolor (L) Moench). However, collecting data on these architectural traits is labor-intensive and time-consuming, especially when using numerous lines in quantitative genetic studies or breeding programs. Therefore, we used the high-throughput field-based robotic platform PhenoBot 1.0 to collect whole canopy stereo images from a large association mapping panel under field conditions. These images were used to create a plot-based 3D reconstruction of the canopy from which phenotypic features were automatically extracted. These features included: plot-based plant height (PPH), plot-based plant width (PPW), plant surface area (PSA), and convex-hull volume (CHV). A small sub-set of sorghum lines were used to obtain groundtruth measurements to validate the image-derived descriptors, and determine their biological significance. PPH was highly correlated with manually measured plant height; PPW correlated with SinAL, defined as leaf length multiplied by the sine of its angle; PSA was associated with manually measured total plant surface area; and CHV was a function of both flag-leaf height and SinAL. Association mapping of PPH identified chromosomal regions containing known plant height genes, confirming the accuracy of the automatic feature extraction process. For the other phenotypic features, significant markers were identified within genomic regions that have been previously reported to control plant architectural characteristics in sorghum such as tiller number, shoot compactness, leaf length, surface area, and angle. The image processing method used in this study contributes new knowledge to the development of high-throughput phenotyping techniques and represents a novel tool for plant breeders.