virtual environments
James Sinnott
Introduction
Other current research on the REVEAL project is looking at methods for automatically recovering the geometry of a real world scene,
through the processing of video sequences.
The work described here is concerned with the next stage: processing and understanding the recovered geometry.
These pages describe one particular component of this work: SQUIDS (SUperquadrics with Interactive DeformationS).
SQUIDS
SQUIDS is a prototype tool that allows the interactive deformation of superquadrics within a virtual environment. This section will explain why SQUIDS was developed, and briefly describe the main components of the SQUIDS deformation interface. This text has been taken from a paper presented at the Eurographics UK Chapter 18th Annual Conference. The full paper can be found here.
Why SQUIDS has been developed
The ultimate aim of this work is to develop a 3D scene understanding system. The input to this system will be a 3D polygon mesh that has been reconstructed from image sequences captured from a real-world scene. The task of the system will be to locate and identify, within this polygon mesh, the various objects that make up the scene.
It is anticipated that a number of different model matching techniques will be used, although the primary approach will be to fit models based on deformable superquadrics to the 3D data.
Superquadrics are a family of simple parametric models that can represent a wide range of 3D forms with just a small number of controlling parameters. Parametric deformations are used to further increase the range and complexity of forms that can be modelled. Complex objects can be modelled as collections of simpler parts, using arrangements of deformed superquadrics.
The images below illustrate some of the different forms that can be modelled with superquadrics (left) and with deformed superquadrics (right).
It is not expected that the system will be fully automated, but that it will take an interactive, user-driven approach. Given the polygon mesh reconstructed from a scene, it is envisaged that the system will operate as follows:
- The user interprets a set of points from the polygonal scene data, deciding what object they represent.
- The user selects an appropriate object model and suggests an approximate initial fit of the model to the data points.
- The system then automatically refines the fit, recovering the precise parameters of the object from the data.
In order to provide the initial fit to the data, the user will need to be able to specify a set of starting parameters for the model. These starting parameters will include the position, orientation, dimensions and any other controlling parameters of the model.
As models based on deformable superquadrics are to be used, then the user will also need to specify the parameters of any appropriate deformations.
There is therefore a need for a simple and intuitive method for applying deformations to superquadrics. The problem is to provide a suitable way for the user to alter the controlling parameters of a deformation.
Some form of deformation interface between the user and the superquadrics is required. By manipulating this interface, the user should be able to alter the controlling parameters of a particular deformation, and then have the deformation applied to the superquadric. A number of requirements have been identified for the interface:
- It must be simple to use within a virtual environment.
- It should operate and respond in real-time
- It must be intuitive: there should be an obvious mapping between the operation of the interface and the resulting deformation. The user should be able to predict the results of their actions.
The SQUIDS deformation interface
The basic principle of the SQUIDS interface is the deformation node.
Deformation nodes are objects within the VE that correspond to the controlling parameters of a particular deformation.
Each node has a path along which it may be moved. A node's position along its designated path determines the value of the corresponding deformation parameter.
By interacting with the nodes, dragging them along their paths, the user can intuitively alter the effect of the deformation on the superquadric.
SQUIDS provides real-time feedback - as the user manipulates the deformation nodes, the superquadric is continuously re-deformed to reflect the new parameter values, and this visual feedback helps to guide his actions.
The follwoing image illustrates the main components of the squids deformations interface:
What SQUIDS supports
SQUIDS allows the user to interactively apply three forms of global deformation to superquadrics:
- Global linear tapering
- Global linear bending
- Global axial twisting
Contact
For further information please contact:
| James Sinnott
Department of Computer Science University of Manchester Oxford Road Manchester M13 9PL, United Kingdom Email: sinnottj@cs.man.ac.uk Phone: +44 (1) 61 275 XXXX |