Résumé
The relationship between foods structural properties and masticatory behaviour has been extensively characterised in literature. Electromyographic and jaw movements recordings are the most common methods used to study mastication. These methods are quite expensive, required preparation of the subjects and constraint the study within a laboratory set-up. Moreover, they are barely applicable for characterising specific population such as children or cognitively impaired subjects because of the non compliance to the protocol. Because of these difficulties, chewing behaviour in these populations has mostly been characterised using video (1,2,3). Video observations were found to be a good alternative to electromyography to determine clinical parameters such as the number of chewing cycles, chewing time and chewing frequency (3). Artificial vision has been used in various areas (from the videogames development to the communication disorder research (4)) for characterising positions and movements. In this study we investigated the applicability of artificial vision for tracking 3-D jaw motion from video recordings, as an alternative to kinematic methods. We also used a stereovision system in an attempt to quantify kinematics parameters: closing duration and velocity, opening duration and velocity, lateral amplitude and vertical amplitude. Main challenges consist to a). Conceive a vision data acquisition system respecting multiple constraints (low cost, convivial man-machine interface, user-friendliness, realist environment of real world applications), b) Determinate if artificial vision allow to automatically detecting and tracking of characteristic facial points from video sequences, c). Establish an image processing and analysis chain to measure and calculate masticatory kinematics parameters with an acceptable precision, and d). Adapt and improve the acquisition protocol for specific population such as children or cognitively impaired subjects. Technically, the vision data acquisition system is completely contactless and wireless (wires are the main inconvenience of the existing optoelectric system); two low cost cameras are used for image capture. In the first time, we position little reflective or colour markers on the interesting facial points of subject. Then marker recognition and tracking is performed. We detect 3D coordinate of each marker using the disparity between two captured images (stereo-vision). Finally, based on 3D movement analysis of markers, masticatory kinematics parameters will be calculated. In order to satisfy required precision, the first step of this study consists in camera calibrating to correct intrinsic radial and tangent distortions (5). The rotation and scale variation introduced by the free placement of subject face will be taken in account by the image processing chain. We make also some comparisons with obtained results by the optoelectric system to test and valid final artificial vision prototype.