Biometrie feature detection from surveillance data using non-calibrated techniques

The development of high quality video surveillance systems provides the ability to perform measurements from image data. This paper describes and demonstrates the extraction of basic human biometric features from surveillance video camera data using an uncalibrated single-view surveillance camera system. Perspective-based photogrammetric techniques are applied in situations where only minimal real world information is available. This information usually includes the reference height and two orthogonal sets of parallel lines on a reference plane. These all must be detectable both in the real world and be in the field-of-view of the camera. Using this type of setup, the orthogonal distance from the reference plane can be computed at any scene point. The measured biometric features include height and body dimensions. Additionally the gait features and the walking profile can be estimated when photogrammetry techniques are extended from a single image to a video stream. Such information can be used for forensic investigation and in various types of other applications where photogrammetric information is needed. This can be done without further information about the camera calibration or position.

I. Introduction

The interest for surveillance camera systems and their applications has found great interest in recent years. Organizations and individuals are looking for more and more automated surveillance camera systems to monitor and save important resources. Nowadays, such systems are targeted to identified individuals and the applications have ended into low level products such as in mobile devices.