english only
Data Analysis
The WearCam Data Analysis involves many steps, depending on the features we want to extract from the WearCam video. Computer Vision algorithms are used to detect and track faces and objects, and statistical analysis is applied to the results. This data can then be used to study the child's visual behaviour (see WearCam & Autism).
Face Detection
The face detection system uses a cascade of weak classifier (Haar wavelets) with AdaBoost to detect faces in the video frames. The position of the face is then compared to focus zones (i.e. the center zone of the image, the edges of the field of vision) to decide whether the kid is actually looking at the person or not.
The system is rotation invariant, to accomodate children who are wearing the camera in free-play situations and may look at people in all different kind of orientations.
Gaze Direction Detection
The forward direction of the field of view of the Wearcam is only an approximation of what the child is actually seeing. The center of attention of the visual field varies significantly when the child is looking sideways. A small mirror placed on the bottom of the wearcam captures the eyes region of the face.
Gaze direction detection is computed using Support Vector Regression. The system creates a mapping between the appearance of the eyes and the direction of the gaze in the image. There is no calibration required by the user, as detection is performed offline.
The system can detect the direction of the gaze with an accuracy of about 2.5° for children and 2° for adult subjects. The image shows a normally developing child wearing the WearCam and playing a touch-screen game used for gaze tracking performance estimation.
Object Detection
Object detection is currently done through simple color and shape detection. A model of the object color is learned from some examples of the object using a mixture of 3D histograms in the YCbCr space. When a new frame (or image) is presented, the response of the object histogram (i.e. how likely it is that a pixel has the same color as the object) is computed for every pixel.
The shape of the object is extracted in terms of roundness and excentricity from the training examples. When the histogram response from the color of the object is obtained, we compute how similar they are to the object model. This allows the system to distinguish between objects with similar color (e.g. a red cup and a red pencil) but different shapes.
Videos
 Face, Gaze & Object Detection Applied to an Interaction Scene |
Selected publications
People involved in this project
