Optimisation the real time implementation of the Viola & Jones face detection algorithm on RISC processor

This article describes the optimization real-time implementation of Viola & Jones face detection algorithm, based adaboost, on RISC processor combined with a genetic approach (called AdaBoost/GA). This algorithm detects the faces in image or video sequences, with rate detection about 97% on the basis of image CMU & MIT. In this work we present a study of the algorithm in terms of complexity and calculation, resources consumption and parallelism. We present a various optimizations used to obtain a rate of treatment, about, 40 images/second, and 320 × 240 pixels of image size. These results were obtained by exploring various techniques of optimization on processor RISC (unwinding of loops, rotation of registers, parallelism, extension SE...) on a processor Pentium IV 2.0 GHz and equipped with 2.0 Giga byte of RAM