Lip tracking program implemented on FPGA

Abstract

The first and crucial stage in audio visual speech recognition (AVSR) system is lip tracking. It is the process of partitioning lip contour from facial video sequences. Many efficient algorithms have been developed to achieve good and accurate segmentation. Most of these have to be performed entirely by software because of their complex computation. In this paper, we present a novel hardware- based lip tracking model, implemented on DE2 FPGA (Field Programmable Gate Array) Board of Altera. Our mathematical tool is the combination of K-Means clustering and shortest path Dijkstra algorithms. The experimental results have proved that our proposed technique is well adapted to face with the presence of beard, teeth and even to lips with weak color contrast. The pipeline design in our hardware structure considerably enhances the processing speed more than 700 FPS) in comparison with software. (FPS: Frames per second).