face-rec-demo.zip_DEMO_PCA OPENCV_face rec_opencv pca

This is a demo of the initial model for Active Face Learning. It "learns" an initial model of a person's face that can be used to bootstrap the process of learning to recognize faces. This initial model is good enough for initial interactions between a user and a social robot. It may mistake others for this person, and may not always recognize this person when present. That's OK. The goal at this stage is simply "good enough" to keep a user interested and interacting so the robot can continue learning. Few face recognition systems are able to do even this much. Typical approaches to face recognition require many example images before meaningful recognition can begin. To use the demo, 1. Unzip all files into one directory 2. Plug in a webcam 3. Launch framecap to enable video feed to the demo 4. After framecap is running, and displaying live video, launch InitialFaceModel and follow the on-screen instructions. When drawing the face ellipse, extend the lower edge of the ellipse a bit below the chin line 5. After InitialFaceModel has finished, close it and launch FaceFinder 6. Select either detection alone, or detection plus tracking. The combination of detection plus tracking improves detection quality. 7. Click Start to recognize using the initial face model For a conceptual introduction to Active Face Learning, see http://www.cognotics.com/vision/facerec/activelearn/index.html For technical details, please see the CVPR workshop paper on this at http://www.robin-hewitt.com/pubs/hewitt-belongie-cvprw06.pdf