Oral Detection And Tracking In Video Images

With the rapid popularity of multimedia and network technology, lip detection and lip tracking has gradually attracted people’s attention; and has been the important part of lip-reading and lip moving synthesis. It can also improve the recognition speed, and in order to compensate the lack of these teachers and improve these children’s self-learning ability, it can help teachers teach the deaf children.The project analyzes the related research on lip feature location and lip tracking, and mainly researches both technologies:one is the deformable templates method which can accurately exact geometrical features of the objects, but has big computation; the other is particle filter which is the common tracking technology, and it has better effect on tracking big objects, while it can lead to point drift for objects which have micro-features. This project uses deformable templates and particle filter method to get lip region features, and has a research on lip feature location and lip tracking for thevideo sequences. Main work is outlined as following:1. Based on these questions that lip location was impacted by the similarity of the lip-color and skin-color of the yellow race, beard, teeth, the uneven illumination and so on, the project gives a lip location method based on CIELab color space. Compared to the existing Red-Exclusion algorithm, CCD algorithm and Chromatic feature extraction algorithm, this algorithm can resolve these problems which contain the dependence on illumination and so on to avoid these noises’impaction on lip location so that it can exact the lip region more accurately.2. Based on these questions which contain the initialization on the deformable template and its multi-parameters’optimization, the project gives a method which uses projection algorithm to locate the lip corners. Have an initialization on deformable template can resolve the problem the complexity of solving process, and it lays the foundation on exacting the lip’s feature points more accurately.3. Based on these problems that there exits lip’s deformation, rotation, scaling, continuity of the edge and so on in the process of pronunciation, the projects presents an algorithm of exacting lip contours based on deformable templates, using multiple curves to approximate the lip contour, and using the edge and trough as the energy function to ensure the lip contour. The experiment shows that the algorithm avoids the impact of the uneven illumination, teeth, tongue and so on. They have an impact on more accurately exacting lip contours, and get the fit parameters reflecting lip changes.4. Based on the question that it is hard to track lip micro-feature points, the project presents a method which uses particle filter to track lip feature points. When using particle filter we must assure through experiments the size of the feature rectangle and the particles’number, so that we can ensure the accuracy and efficiency of tracking. And through experiments the project makes a statistic on the inaccuracy of tracking, adjusts the lip regions, and reduces the location deviation. The experiment shows that the algorithm has an advantage on the accuracy and efficiency of tracking.In conclusion, it gives a study about lip detection and tracking for the video sequences. Some experimental verification are done for key technologies, achieving desired result.