Second Interim Technical Report for Advanced Research in Range Image Interpretation for Automated Mail Handling
Files
Penn collection
General Robotics, Automation, Sensing and Perception Laboratory
Degree type
Discipline
Subject
Funder
Grant number
License
Copyright date
Distributor
Related resources
Author
Contributor
Abstract
This is the report on range image interpretation of singulated irregular parcels. The overall approach to this problem was to find the most general and generic method that will interpret the data yet is computationally economical. We have observed that the large majority of objects that are classified as Irregular Parcels are convex or can be modeled as such plus some deformation. Hence the primitive model that we have chosen, is the superquadric plus deformations such as tapering and bending along the major axis. The superquadric model is an analytic representation of volume for which cross-sections are a class of curves varying between rectangular to elliptical (of course, circular and square shapes are included). As a side product of this representation we obtain naturally the position, orientation and the scale of the object. The principle of recognition is a fitting procedure which changes the parameters so as to minimize the difference between the data and the volume of the model. We believe that this approach is more general than any previously model based approach including CAD/CAM systems. In comparison to the Generalized Cylinder model proposed by Binford [3] and continued with his students [13], our model has the advantages that: a) It does not require any preprocessing, that is the fitting is performed on raw 3-D data points. b) The interpretation is performed in coarse to fine fashion, or from global to local.