Left item detection is a key issue in security
systems based on Computer Vision. These surveillance systems can be implemented
in many different ways, most often used ones are the real-time video
surveillance systems. These systems usually rely on simple video streams
captured by surveillance cameras, which in some situations could not perform as
good as expected. A new technology recently released to the public, the
Microsoft Kinect sensor, through its capability of capturing RGB-D data
theoretically can give a much more precise and robust approach for detecting
objects and humans in a scene. In this project we try to demonstrate the
possibility to build such kind of a system around your own sensor in your own
home.  The lack of time and several setbacks forced us to simplify our task and reformulate our research. We used an algorithm for human detection which was proposed by Matteo Munaro etal. It features a novel depth-based sub-clustering method explicitly designed for human detection, this multi-people tracking algorithm assumes that people are moving on a ground plane thus it is able to track them robustly without needing to scan the whole frame. The next step would be: suppression of the detected people on images and detecting the luggage or item's that the human's are leaving after in an RGB-D data. For more information about the approach we would like to refer you to the next pages of our website and the flowchart below that is providing our approach.
|
 |
|
|
|
