Implementation And Research Of Embedded Face Detection Using Adaboost

Face detection technology has many important applications in many situations. For example: face recognition based on video streams at airport, security area, index on digital figures. During the past ten years, face detection is the most challenged problems at image processing area. With the study progressed, many algorithms has been published. In 2001, Viola and Jone has introduced an important algorithm that is called fast face detection based on Adaboost. At PC platform, they have achieved real-time detection rate. But, can this algorithm achieve same detection rate at embedded system?Staring from software implementation, two detection mechanisms have been studied. One is called classifier magnifying, the other is called image sampling. After showing the advantage and disadvantage of the two mechanisms, the detection rate at PC platform has been compared. The image sampling mechanism is 3 times slower than the classifier magnify mechanism. Then, after migration to the embedded system based on PowerPC 405 of Xilinx XUP development board, three optimization methods have been raised. They are float-point to fix-point conversion,parameter modifier and acceleration with co-processor. After using these three methods, the detection interval of the image sampling mechanism is 0.08s compared to 0.16s of classifier magnify. After that, an ASIC implementation of face detection has been raised, and C model has been constructed. And the full-hardware implementation base on FPGA achieves 0.02s to detect face.