Advanced Imaging


Advanced Imaging Magazine

Updated: January 12th, 2011 10:01 AM CDT

Security in 3D

Overcoming problems of lighting changes, partial obstructions and multiple targets
Rockefeller Center observation deck
Changing lighting conditions on the Rockefeller Center observation deck provide numerous tracking challenges that were solved by a 3D system.
tracking people
Tracking a person in an area of bright sunlight, such as this office setting, would be difficult in 2D.

By Barry Hochfelder

3D imaging isn't affected by constant changes in lighting and apparent color, requiring that the subject only be partially visible to a single stereo camera to be correctly tracked; multiple cameras are used to extend the system's operational footprint and to contend with heavy occlusion. In addition, tracking with a 3D stereo camera can provide the precise location and movement of each individual, even in a crowded environment.

PersonTrack provides an API (Application Programming Interface) that includes real-time updates of all people's locations in the facility, person counts for the full facility or selected sub-regions, as well as alerts and counts for people present in specified restricted areas.


Stereo cameras provide an additional dimension to the data, depth, or distance from scene to camera, which can be used to greatly improve performance. Depth data is relatively robust to lighting changes and can handle occlusion explicitly. It also provides metric sizes of objects to avoid the scaling problem that occurs in 2D-based tracking. The baseline length is a key parameter when comparing stereo approaches. Wide baseline systems provide accurate depth when correct correspondences are found, but suffer from more occlusion and substantial changes of viewpoint and thus may give less dense depth data and encounter more false matches where cameras in stereo pairs are separated by meters.

For tracking applications, PersonTracker has several advantages over systems based on wide baseline stereo. Unlike most wide baseline systems, it does not assume that people are moving on a ground plane. Person detection, matching and occlusion handling in the tracking algorithm are directly based on dense and accurate 3D information, which makes those modules easier to implement and more reliable. Segmentation based on depth data can be used to improve appearance models of each person.

The G2 is a PowerPC-based computer that runs Linux. At its component level, the G2 consists of an embedded PowerPC chip (666 MHz 440GX) connected to two CMOS imagers, a TYZX DeepSea II chip, an Analog Devices Blackfin DSP (Digital Signal Processor) and some memories using an FPGA. It is designed with a particular dataflow architecture/FPGA configuration that allows the various components to be configured at a course level of granularity. This allows the system to be applied to distinct tasks without modifying the firmware in the FPGA.

Subscribe to our RSS Feeds