A Robust Non-cooperative Face Recognition
Scheme from a Distance in Maritime Domain
Biometrics applications are becoming of particular importance
in the maritime domain such as informing decisions involved in interception
operations and monitoring of watercraft traffic in critical littoral areas of
operation. It is of particular interest for Navy to acquire biometrics of
persons of interest in small or medium crafts, using self-contained shipboard
or patrol vessel mounted systems at operational distances of 20m to 100m. Considering the operational distances of
20m-100m, face recognition becomes a more favorable biometric when compared to
iris recognition. There has been considerable work in face recognition area for
human recognition at a distance. However, most of the existing face recognition
systems require a cooperative individual and often under a controlled
environment. While these techniques are much more capable of identifying
cooperative subjects, they are almost entirely incapable of identifying
non-cooperative subjects for applications such as surveillance, where the
observed individuals are non-cooperating and non-habituated. Non-cooperative
subjects are people who are either unaware that a biometric system is in place
or make no effort to either be recognized or to avoid recognition.
Through this research, we implemented a new
multi-people context-based detection and tracking algorithm that can reliably
track people and maintain theirs IDs from distance under realistic illumination
conditions and with various occlusions.
From the detected and tracked human body, we then implemented a new face detection and tracking algorithm using case-based
reasoning paradigm. The technique can
detect and track multiple faces in real time and under different face poses
with significant camera motion. A local
patch based facial recognition algorithm is then implemented that can perform
robust and real time facial recognition.
These techniques were validated extensively under various outdoor
conditions, different distances, different camera motions, and different
illumination conditions (day and night) for many different subjects. These experiments demonstrate the
feasibility of our methods for non-cooperative face recognition from distance
in maritime domain.
Video demos for face detection and tracking may be
found here.
Video demos for face detection, tracking, and
recognition may be found here.