Yield component traits such as plant height and stem diameter are dominant phenotypic data for biomass sorghum yield prediction. Extraction of these traits by machine vision during the growing season significantly reduces labor and time cost for large breeding programs. An automated 3D point cloud processing pipeline was developed to quantify different phenotypic variations in plant architecture of infield biomass sorghum. The input point cloud was generated by three side-view stereo camera heads placed vertically to capture extremely high plants. The features were extracted on a row plot basis instead of individual due to severe occlusion caused by densely populated leaves. Available features include plant height, plant width, vegetation volume index, and vegetation area index. Our strategy was to slice the point cloud along row direction into several equal volume slices and sum up the feature values with weights based on the point population and distribution in each volume slice. Therefore, the results were robust against empty space and abnormal individuals in the row plot. In addition, a semi-automated user interface was developed for users to measure stem diameters from the stereo images according to their specific sampling strategies. Users only need to zoom in on a stem segment and pick four corners of the rectangular segment. Metric measurement is then computed automatically based on image patch stereo matching using normalized cross correlation. The extracted stem diameters were compared to manual measurements in the field and a high correlation was obtained. The extracted features revealed great potential for automated field-based high-throughput phenotyping for plant architecture.