In Ice-Hockey, performance of the player’s shots depends on their skill level, body strength as
well as stick’s construction and stiffness. In fact, research suggests that one of the primary reasons
that the elite players generate much faster shots is their ability to flex their hockey stick. Thus,
reconstructing the deformable 3D shape of the stick during the course of a player shot has important
applications in performance analysis of ice-hockey stick.
We present a new, low cost, portable system to acquire videos of a player shot and to automatically reconstruct the stick shape’s deformation in 3D. This thesis is a sub-part and contributes
towards the ultimate goal of the pipeline in many different ways. First, designing a mobile stereovision setup and its calibration, capturing a lot of data acquisitions with different players shooting
in different styles. Second, developing a two step pruning methodology to prune structurally thin
and fast moving ice-hockey stick from noisy reconstructed point cloud in 3D. Third, automating the
process of initial rigid alignment of the stick template in the noisy reconstruction. Forth, reducing
the effect of noise by using medial axis approximation approach and suppressing the hand occlusion
effect on the final template bending by a curve fitting approach. This pipeline is also robust against
different ice-hockey sticks along with different players, shooting at different styles.