Quantifying how motion flows from one body segment to another is a long–standing challenge in sports biomechanics, rehabilitation, and human–robot interaction, especially for non-periodic movements such as over-arm pitching. We introduce a convolution–based dissimilarity metric that treats each pair of time-series as the input and output of a Linear Time-Invariant (LTI) system. Replacing the scalar inner product with the geometric product yields a framework that is agnostic to dimensionality: for planar motions the convolution coefficients form a vector in \(\mathbb {C}^{M+N-1}\) , whereas for spatial motions they inhabit \(\mathbb {H}^{\,M+N-1}\) . Similarity is then quantified with the complex or quaternion \(L_{2}\) norm, preserving scale invariance while embedding rich directional information. Experiments on video/motion-capture data of baseball pitching motions show that the proposed metric (i) discriminates subtle ankle-to-wrist coordination patterns, and (ii) remains stable under camera shooting angle in horizontal plane of \(\pm 40^{\circ }\) . These findings establish a direct bridge between classical LTI analysis and Geometric Vector Time-Series (GVTS), suggesting a unified language for multi-modal sensor fusion, explainable graph neural networks, and real-time feedback applications.