The Viability of Image Registration as a Method for the Quantification of Displacement in Penetrating Impact Experiments

Experimental characterisation of tissue deformations associated with penetrating impact of fragments from explosive devices is challenging. Whereas experiments involving ballistic gelatine tissue simulants enable direct visualisation of deformation patterns, quantification of these deformations remains difficult. This thesis investigates the use of image registration for this purpose.
Image registration methods optimise alignment of corresponding structures in image pairs, and in the process estimate the deformation fields that best achieve this. In the current context, it is hypothesised that registration of consecutive images from videos of gelatine penetration events can enable the corresponding gelatine deformation fields to be estimated.
Three main activities were undertaken towards validation of this hypothesis: the proposed registration approach was tested on a series of synthetic images emulating the types of deformations expected in penetration events; the approach was then tested on images derived from a carefully controlled indentation experiment, in which a block of gelatine was deformed quasi-statically with a rigid indenter while the resulting deformation was filmed; and finally it was tested on video footage from projectile penetration experiments, in which metal projectiles were fired into blocks of gelatine and filmed with a high speed video camera.
A series of complementary studies was also undertaken in support of these experiments. Firstly, to better understand the parameters of real penetration scenarios, the fragment generation and flight behaviour of a typical explosive device were analysed. Secondly, to improve understanding of the material behaviour of the test gelatine, mechanical characterisation tests were undertaken, and a visco-hyperelastic constitutive model was proposed.
The individual registration operations themselves appeared to perform well, in the sense that initially disparate consecutive image pairs were brought into good alignment. However, composition of the corresponding transformation fields, necessary for tracking accumulated deformations over the course of a video sequence, was found to yield artefacts and unphysical deformation estimates in some cases. These were judged to result both from deficiencies in the methods themselves, and flaws in the experimental arrangements. Therefore, while the proposed registration approach appears to show promise, further work is needed to establish its validity conclusively. The thesis closes with a discussion of possible approaches to the latter.