A minimally intrusive, vision-based, computational force sensor for elastic
ally deformable objects is proposed in this paper. Estimating forces from t
he visually measured displacements is straightforward in the case of the li
near problem of small displacements, but not in the case of the large displ
acements where geometric non-linearities must be taken into account. From t
he images of the object taken before and after the deformation. we compute
the deformation gradients and logarithmic strains. Using the stress-strain
relationships for the material, we compute the Cauchy's stresses and from t
his we estimate the locations and magnitudes of the external forces that ca
used the deformation. A sensitivity analysis is performed to examine the ef
fect of small deviations in the experimentally captured displacements on th
e estimated external forces. This analysis showed that the small-strain cas
e is more sensitive and prone to numerical errors than the large-strain cas
e, Additionally, a related method that is indirect and iterative is also pr
esented in which we assume that we know the locations of the external force
s. Numerical and experimental studies are presented for both micro- and mac
ro-scale objects. The main conclusion of this work is that the vision-based
force estimation is viable if the displacements of the deforming object ca
n be captured accurately. (C) 2001 Elsevier Science B.V. All rights reserve
d.