Statistical gait description via temporal moments

Statistical recognition techniques have already been shown to achieve good performance in automatic gait recognition. However, the metrics were only statistical in nature and did not describe the intimate nature of gait. Accordingly, new velocity moments have been developed to describe an object and its motion throughout an image sequence. These moments are an extended form of centralised moments and compute descriptions of the object and its behaviour evaluation shows that the velocity moments have the required descriptive capability and analysis on synthetic imagery shows that the velocity moments are less sensitive to noise than an averaged comparator moment. This is largely due to the integration of data from the whole sequence. An extraction procedure has been developed to find moving human subjects and we are currently evaluating the performance of this promising new approach in automatic gait recognition.