Image recognition work for the UBC Agrobot Design Team

Biomass Estimation using Intel Realsense Depth Camera

A part of the contest involves estimating the amount of biomass present in a crop field in mass. The biomass team and I decided that using an Intel Realsense Depth is the most appropriate. I wrote a script that processes .bag files into rgbd data. The depth of each pixel is then used to estimate the area -> volume -> mass of the vegetation. The script is still in development, but the results are promising.

Plant Image Segmentation

Implemented an algorithm that efficiently segments vegetation/plants from the rest of the image, such that our object detection folks can use the segmented plant data to train a more generalized model. The process included the use of Otsu thresholding, and various filters. The algorithm was implemented in Python and OpenCV.