Design And Implementation Of Embedded DSP-Based Face Detection System

Target detection is has been a focus of research in computer vision and pattern recognition, after decades of development, there have been new breakthroughs in the theory and algorithms, detection accuracy and efficiency are constantly improved. Face recognition is an important target detection direction, after years of theoretical and breakthrough technologies in the field of technology is relatively mature, recognition accuracy and rate the industry has come to the practical level. Especially in video surveillance, and identity certification in the field, face recognition is important, as more integrated, on demand for the functionality embedded products more widely.Face detection can be achieved through face location, in order to make the appropriate positioning of face recognition feature. Adaboost algorithm is the classic face detection algorithm, which uses a "boosted" thinking, in the premise of ensuring the detection rate, detection speed very quickly, especially in the real-time requirements for the high scenario, Adaboost algorithm can take into account both compared high accuracy, but also has a high detection efficiency. Therefore, we select the Adaboost algorithm as the content of the study. DSP developed in the last decade as a dedicated embedded CPU has great broad market, on the one hand it can meet the operational requirements for the professional, on the other hand compared to general-purpose CPU, it has a relatively low cost. At present in the field of video surveillance, DSP applications particularly widespread, on the other hand TI DM6437screen provides a rich interface with a corresponding image processing LIB library, can be easily embedded video processing system development, this paper select Embedded DM6437as an experimental platform. Based on the study, through the familiar face detection technology, read a lot of literature, after a lot of testing and coding, and ultimately in the TI DM6437platform to achieve a complete system based on Adaboost face detection algorithm, and for that done a lot of optimization algorithms, and ultimately makes the algorithm running in the DSP platform performance to meet the overall requirements of real-time, this study face detection system for subsequent embedded face recognition system has a certain reference value, with a strong practical value.