A Virtual Curtain for the Detection of Humans and Access Control - Advanced Concepts for Intelligent Vision Systems - page 98

deliberately chose to rely of laser scanners instead because they can be directly
embedded in a door frame and do not require a controlled environment. In
addition, the complete solution is required to operate in real time with a limited
amount of processing power, and it appeared that the choice of laser scanners
proved adequate, retrospectively.
The article by Sequeira et al . [2] is representative for the problems faced with
laser scanners. Some of them are:
- occlusions and shadows. Objects located beyond other objects are not de-
tected (occlusion) and, likewise, it is impossible to interpret the scene beyond
the first object reached by the rays of a laser (shadowing).
- angle of acquisition. Laser scanners have a narrow aperture angle in one
or two dimensions. Also, as the technology is based on a sender/receiver
mechanism, physical properties and the angle of incidence are important.
- scan overlap. To interpret fast movements, data has to be captured either
at a high speed, or with a fair overlap between successive scans. A practical
solution consists in the use of linear scanners with a high acquisition rate.
- scan resolution. Radial scanners output distances of closest objects for a
fixed set of angles. The radial sampling might be uniform, this does not
mean that the precision on the distance both in the direction of the light
ray or in the direction perpendicular to it is uniform as well. In fact, objects
should be closer to increase the measurement accuracy, but the price to pay
is an increased shadowing effect.
The paper is organized as follows. As we propose an original placement of laser
scanners and a new method to build a virtual curtain (which is an invisible and
immaterial membrane), we first describe our set-up in Section 2. From this ar-
rangement of scanners, we derive the notion of a virtual curtain, described in
Section 2.2. This concept is the key of a classification process detailed in Sec-
tion 3. First, surfacic features are extracted from the intersection of the curtain
and an object or a person that crosses it. Then we concatenate these features
over time to derive a windowed temporal signature. This signature is then used
to identify the scene by a classification process; the purpose is to raise an alarm
when the normal situation of a single person crossing the curtain is not met, for
example when several persons want to pass the door simultaneously. Results of
this classification method obtained over a database of more than 800 sequences
are provided in Section 4. Section 5 concludes the paper.
2 Original Set-Up
A real security application requires that the system is insensitive to lighting
conditions. Consequently, we cannot afford using a background subtracted video
stream to recover the binary silhouettes of the walkers. Instead, we use laser
devices described in the next section. Then we develop the concept of virtual
curtain.