Estimating material properties of deformable objects by considering global object behavior in video streams

Abstract

One of the crucial components in improving simulation quality in physics-based animation of deformable object is finding proper material properties that define the movement upon external excitation. Most work in the estimation of material properties for highly deformable objects involves applying localized force to a point on the object's surface with mechanical devices and measuring the displacement of the surface at the contact point and surrounding points. While understanding this localized behavior provides a step towards accurately simulating objects with known material properties, an understanding of the global behavior of the object undergoing deformation is more important for many practical applications. This paper describes both the computer vision based techniques for tracking global position information of moving deformable objects from a video stream and the optimization routine for estimating the elasticity parameters of a mass-spring simulation. The collected data is the object's surface node position of object over time which is used to a data-driven simulation of that object to match the behavior of a virtual object to the corresponding real one. This paper demonstrates that estimating material properties of highly elastic objects by matching the global behavior of the object in a video is possible with the proposed method and the experimental results show that the captured and simulated motions are well matched each other.