Registration of hand-grip pattern in smart gun

Secure Grip1 focuses on the development of a hand-grip pattern recognition system, as part of a smart gun. It is intended for use by the police. We collected the hand-grip data from police officers in three sessions in a row with time intervals in between. If data for training and testing come from different sessions, however, the verification performance is not satisfactory. This is caused by the variations in data collected from different sessions of the same subject. Among others, one type of across-session variation is the hand shifts of subjects. Therefore we applied templatematching registration (TMR) as a data-preprocessing step. This has helped to reduce the Equal-Error Rate (EER) of the system by 1.7%. Besides, we discovered that the hand shifts of the same subject are on average smaller than those between different subjects. This has inspired the implementation of a new classifier, based on both the grip patterns and the hand shifts. With the new classifier the EER of the system has been further reduced by 3.0%.