We demonstrate a novel approach to modelling arbitrary temporally-deforming objects using spatio-temporal Fourier descriptors. This is a continuous boundary descriptor, which can handle shapes that vary in a periodic manner (such as a walking subject). As such, we can handle non-rigid, moving shapes that self-occlude. We show how this approach has led to successful shape extraction and description with both laboratory-sourced and real-world data. A consequence of exploiting temporal shape correlation in this approach has led to very good tolerance of noise and other positive performance factors. Further to this, our new approach holds sufficient descriptive power not only for extraction, but also for description purposes, and we have been pleased to note high recognition rates in human gait recognition on a large database.