An Assessment on the Feasibility of Footsteps as a Biometric

点击翻译按钮获取中文摘要

A variety of established biometric technologies exist, many supported by the EPSRC's Think Crime portfolio. In the last two years these include the recognition of people from their voices, the recognition of faces from images and video sequences, handwriting recognition and finger print recognition, all of which have been identified by the Police Information Technology Organisation (PITO) as having 'clear police applications'. Another researched biometric is gait, the way a person walks. Gait recognition is described by PITO as a relatively embryonic biometric but one which has potential uses in police surveillance applications should the technology mature. The aim of this research programme is to assess the feasibility of footsteps as a biometric. As a potential biometric footstep recognition has received very little attention as reflected by the open literature and has not previously been considered by PITO as a means of identification. The benefit of the footstep biometric in terms of covert, unobtrusive security is the impetus behind this research proposal. An additional motivation lies in the potential for augmenting gait where footsteps could act in a supportive role, assisting in feature extraction and modelling.The sensor arrangements to capture the footstep signature are seen as absolutely critical to this proposal. In this respect the work is particularly well placed. A local SME, Flooring Technologies Ltd has developed a recycled, rubber based commercial floor tile product. A recent initiative has proven the potential of embedding a piezoelectric sensor that can indicate transient pressures. The envisaged application considered by the company to date is monitoring 'people flow' in given environments: examples include super markets and sports venues. In a shopping mall it might be possible to detect rates of flow at different times of the day for commercial and safety reasons. The classification task is then relatively simple, i.e. human/non-human. Additional value might be obtained if it were possible to deduce more detailed statistics, for example the percentage of customers that are adult or minor, or male or female. Thus footstep 'signatures' could be classified into increasingly higher levels of detail and ultimately, as is the objective of this research programme, to a given person, i.e. as a biometric.The footstep biometric is relatively new and there are a number of adventurous research challenges to be addressed. For the highest accuracies all behavioural biometrics require extensive training data. This is necessary to realise person specific, discriminative statistical models capable of coping with intra-person variations within the general world of inter-person variations. Obtaining these statistics form the heart of this research programme: the features, classifiers and normalisation strategies to enhance discrimination and reduce the influence of extraneous factors. These factors tend to demand significant databases to span the general world of signals plus sufficient enrolment data for any given class. Thus this research programme is to collect what is believed to be the first footstep database of a significant size (~100 persons with multiple client sessions) and will thus produce a meaningful assessment on the feasibility of footsteps as biometric.At this stage it is impossible to predict accuracy and accuracy is a key figure of merit which will inevitably influence viable applications. Thus given the adventurous, high risk nature of this research proposal, funding is sought to enable a 12 month feasibility study. The intention is to answer questions relating to the implementation, accuracy and commercial applications of a footstep-based biometric. This 12 month feasibility study will fund the necessary resources to produce a scientific assessment on the potential of footsteps as a biometric and is an outstanding opportunity for the UK to gain an early advantage in a new biometric.