A Design Methodology for Selection and Placement of
Sensors in Multimedia Surveillance Systems
Aim
Designing to reduce cost for
specified performance
Approach
- Define a performance metric (η)
- Find all combinations s.t. η > required
performance (σ)
- Choose the combination with least cost
We define η as the average probability of capturing the frontal
part of a symmetric object at a particular time instant
Results overview
¾
We
found that 2 cameras in a square region should be placed at diagonals. Center
is not always the best choice.
¾
Similar
(though not same) results were found in irregular areas. The basic idea remains
to get the cameras to cover maximum possible area independently while
minimizing their commonly focused area.
¾
The
sensor performance saturates at certain points. e.g.
We found that adding more sensors beyond 12 or increasing the camera
field-of-view above 45 degree did not increase the performance anymore in the studied
setup. Hence, we can find such threshold values and ensure that we do not spend
money on enhancing the sensors beyond these values.
¾
Actual
tracking results corroborated well with the theoretical optimal camera
placement positions. Tracking provided 69% frames with frontal face information
when cameras were placed at the diagonals.
¾
Use
of motion sensors helped in recovering from situations where the object was
lost. e.g. From (g) to (h) in figure below. Hence
motion sensors can help in improving the performance.
Publication
Siva Rama K. G., Kalpathi Ramakrishnan, Pradeep K. Atrey,
Vivek K. Singh and Mohan S. Kankanhalli. . A design methodology for selection
and placement of sensors in multimedia surveillance systems. The 4th ACM International Workshop on Video Surveillance
and Sensor Networks in conjunction with ACM Multimedia 2006 (ACM
VSSN'06), October 27, 2006, Santa Barbara, California, USA. (To
appear)