5 Conclusions
We introduced a contour-based shape metric and a framework for vision-based OCSC with focus on texture-less objects. Experiments that validate the approach illustrate how energy \(E_{\mathrm{MSN}}\) achieves a balance between reducing the amount of deformation and faster performance, while \(E_{\mathrm{OC}}\) enables more conservative (less aggressive) shape evolutions but presents a slightly slower performance. Some interesting future research lines include testing the performance of \(E_{\mathrm{MSN}}\) on 3D contour data, thus allowing to define target shapes in more detail. However, regardless of 2D or 3D contour data being used, there may be cases where the visual contour does not change significantly. For instance, the object may be undergoing large deformations that cannot be noticed (e.g. a bulging object). This problem could be tackled by extending the method for the analysis of surfaces (rather than contours), with the main challenges being the computation of elastic maps between surfaces as well as the surface parameterisation and the definition of the local frames of reference \(\Re(s)\). Another avenue is to investigate how force information obtained from force sensors on the manipulators (Süberkrüb et al., 2022) can complement \(E_{\mathrm{OC}}\).