Using Hidden Markov Models for Biometric Gait Recognition
Abstract
The need for secure authentication to mobile devices is rapidly increasing with the advent of new technologies. Many of the new mobile devices can be used for various purposes such as internet access, mobile banking, calender etc. As a result of this, sensitive information like phone numbers, address contacts and even financial information are stored on these devices. When valuable information like this is present, it raises serious concerns in case the device is lost or stolen. For protection of the device contents, this thesis proposes a biometric gait recognition method based on the accelerometer data obtained from the mobile device. This method offers an unobtrusive and hence user friendly way for authentication on mobile devices. Biometric gait recognition based on accelerometer data is still a new field of research. Most of the existing methods use dedicated accelerometers to collect gait data and then use a suitable cycle extraction method to extract the gait features. As cycle extraction methods are sensitive towards irregular cycles, this often affects the error rates of such systems. In this thesis, Hidden Markov Models are used for biometric gait recognition. These have already been successfully implemented on various commercial speaker recognition systems, but have never been used for biometric gait recognition. The advantage of this method is that no error-prone cycle extraction has to be performed, but the accelerometer data can be directly used to construct the model and thus form the basis for successful recognition. Two major experiments were conducted: using different preprocessed data sets and using different model topologies, for training and testing the Hidden Markov models. By using the accelerometer data obtained from mobile phones, a false rejection rate (FRR) of 10.42% at a false acceptance rate (FAR) of 9.31% was obtained.