The goal of this project is to “count the Duplos on the table.” The tools available include a point cloud-‐producing sensor (the Kinect), ROS, the Robotic Operating system from Willow Garage, and PCL, the Point Cloud library. An end goal for this project is to provide some usual information to another agent such as a robotic arm motion planner. The information provided could help the motion planner manipulate LEGO Duplos.
Two methods of counting Duplos are presented here. The first method is very naïve and separates a point cloud into clusters of bricks only by color and distance segmentation. It will be shown that this method is very reliable and very fast in a controlled situation. However, this method is shown to be imprecise. The second method improves upon the first by trying to fit known models of Duplo blocks to the point clouds clusters. With a known model is aligned to a target cloud, its shape and full 6DOF can be determined. While much more descriptive in its results, this method is slower and requires training the model before use.
Full report and presentation below.