Monday, September 5, 2011

Tracking, localization and recognition of users in a SmartSpace

One of the group recent activities is in Computer Vision area. It is being developed by two Computer Science students of ''Universidade de Brasilia'': Danilo Avila (me) and Tales Porto.
Nowadays, I'm working at "SEA Tecnologia" developing web systems. Tales works at "Mirante Tecnologia", also developing web systems. We joined this research group at the beginning of this year (2011).

Danilo ÁvilaTales Porto

At the Unbquitous group we're trying to build a system to track, localize and identify people in our SmartSpace called LAICO. We're using some open source libraries like OpenCV, OpenGL and OpenNI and Kinect Sensor from Microsoft as a camera device.
Basically we use the kinect depth data and OpenNI library and drivers to identify new users in the scene and track them. When a new user enters the scene an event is generated and a callback function handles it. This callback function gets the new user's pixels, creates a new rgb image with kinect rgb data and transfers this new image to a recognition module.


User tracked and identified.Kinect.

The recognition module receives the user's image and perform face recognition. Our algorithm uses viola-jones method for face detection and eigenfaces algorithm for face recognition, both implemented by OpenCV library. When the recognition is performed, we return the name of the user identified and the confidence of the recognition back to the tracker module.
In spite of performing recognition in just new users, we keep trying to recognize users already recognized to improve the confidence in the recognition. Then we add a label to the user with it's name and confidence.
At this moment the system can identify and track users in the smartspace with no problem. But there is some issues that can't be resolved using just one kinect, like obstruction and when the user's face can't be get by the kinect because of it's position.
Now we are trying to improve the recognition confidence and to integrate our system with the middleware UbiquitOS.

No comments:

Post a Comment