A pyroelectric biometric sensor system for human identification
The pyroelectric infrared (PIR) sensor makes possible high-performance IR radiation detection at room temperature, while cost and low power consumption make it attractive for security applications. Tracking human targets with such a system has been described,1 but little attention to date has been paid to walking, which can also be employed for purposes of identification and scene surveillance in security applications. It can also be used for tracking multiple persons.
When humans walk, the motion of various parts of the body, including the torso, arms, and legs, produces a characteristic signature. Much work on motion analysis as a behavioral biometric has used video cameras to stream large amounts of data from which the identity of the person of interest can be extracted in a computationally expensive way.2,3 A continuous-wave radar has been developed to record the signature corresponding to the walking human gait.4 We propose a new method by which the features of motion are represented by the processed content of the temporal signal, generated by humans crossing the field of view (FOV) of the PIR sensor module.
Feature representation is key to biometric recognition. From a thermal perspective, each person represents a distributed IR source, the distribution function of which is determined by shape and IR emissivity of the skin at every point. Combined with idiosyncrasies of carriage, heat will uniquely impact a surrounding sensor field, even while the subject follows a prescribed path. By measuring the response thus generated within the FOV of a sensor module, we can model data to create a code vector that uniquely identifies the person.
We have developed two PIR feature-generating sensor systems.5,6 One system is analog, the other digital, and both are derived from the signals generated by humans crossing detection areas. The experimental setup used for analog feature generation is shown in Figure 1. A module containing a pyroelectric IR sensor and a Fresnel lens array was mounted on a pillar to detect IR radiation from the target. Data was collected as individuals walked back and forth along a straight path perpendicular to the sensor. With heat flow, electric charge built up on the sensing element by virtue of pyroelectricity. The resulting charge translated into a current that a current-to-voltage transductance amplifier converted to a voltage signal. Spectral techniques analyzed this signal to extract the individual motion features.
Temporal voltage signals generated by two different individuals crossing the sensor FOV are charted in Figure 2, and the corresponding Fourier spectra, shown in Figure 3, are clearly distinctive.
For the digital feature system, the sensor module in Figure 1 was replaced by a module containing four PIR detectors and Fresnel lens arrays covered with periodic sampling masks, as shown in Figure 4. The binary event sequence was chosen as the digital human-motion feature, with an event defined as thermal flux collected by a pyroelectric detector exceeding a given threshold. The event signal is associated with specific motions of human subjects, such as crossing one or more adjacent detection regions. Figure 5 illustrates two 4-bit digital features (event index sequences) generated by two persons. Again, the individual features are markedly distinct.
During the training stage, analog spectra and digital sequences are modeled in linear regression models and hidden Markov models, respectively, for human recognition in two modalities, path-dependent and path-independent. Average identification rates for a group of 10 persons achieved 90% and 78%, respectively. While the recognition rate can be improved, multiple sensor nodes and additional feature representations may be expected to make the system robust.
