Velocity moments for holistic shape description of temporal features

The increasing interest in processing sequences of images (rather than single ones) motivates development of techniques for sequence-based object analysis and description. Accordingly, new velocity moments have been developed to describe an object, not only by its shape but also by its motion through an image sequence. These moments are an extended form of centralised moments and compute statistical descriptions of the object and its behaviour. Two variations of this new technique are presented. The first uses the non-orthogonal Cartesian basis, while the second utilises the orthogonal Zernike one. Despite their difference in basis, both techniques exhibit favourable characteristics. Evaluation illustrates the advantages of using a complete image sequence (over single images), exploiting temporal correlation to improve a shape's statistical description, while also improving the performance of these statistical features under less favourable application scenarios, including occlusion and noise. To further characterise the velocity moments, they have been applied to gait recognition - a potential new biometric. Good recognition results have been achieved using relatively few features and basic feature selection and classification techniques. However, the prime aim of this new technique is to allow the generation of statistical features which encode shape and motion information, with generic application capability. Theoretical and applied analyses show the potential of this new sequence-based statistical technique and highlight the consistency of its performance attributes with those of conventional moments.