Face detecting algorithm of the Cascade Adaboost on DSP

This paper implements the real-time face detection system with the Cascade Adaboost algorithm on the DSP platform (TMS320DM64_x). Aiming at the feature extraction of the algorithm and cascade multiple classifiers, algorithm level optimization and implementation level optimization are proposed on DSP, such as floating-point to fixed-point Conversion (FFC), loop unrolling, software pipeline and linear assemble language optimization. The experimental result indicates that the system can detect any number of faces with any size at any position in a 320×240 image at a speed of 15 fps. The detection system can achieve a detection rate of 89.5%. The real-time face detection is implemented.