The Design And Implementation Of Face Detection Algorithm Based On FPGA Chip

With the unremitting research of face detection technology in recent years, face detection technology has been greatly developed, and it gradually becomes an independent research project. People are often the most suitable objects which should be monitored in most video surveillance systems, and people's face owns the most important features to be identified for a person.Therefore, face detection occupies an important position in digital control system. Adaboost algorithm is an important milestone in the study of face detection, especially for real-time face detection.Theoretically, using enough samples and rectangle features, the classifier obtained from Adaboost algorithm can get a error rate which infinitely tends to zero through sufficient training. Due to the limitations of the equipment for training, the time to get a classifier on PC often takes several days or even dozens of days.Therefore, it is particularly important to reduce the training time of the classification.Based on the basic principles of Adaboost algorithm, this thesis designed and realized the Adaboost algorithm on the ISE platform. In the training process of Adaboost algorithm, the computation is very large, mainly in obtaining the features and training the best weak classifier. This paper takes full advantages of FPGA's ability of parallel computing and block RAM resources to quickly obtain the values of the sample features and achieves the rapid training of the classifier. Because of the full use of hardware advantage of FPGA.compared to the traditional PC, the training time is greatly reduced. In addition, with the classifier trained in this paper, the detection algorithm is realized on FPGA by the multi-scale detection methods.This thesis introduces the significance of this research and current research at home and abroad first.and introduces the theory of Adaboost algorithm briefly. Then, it gives a very detailed account of the specific method and process in the training of Adaboost algorithm and testing part. The simulation results in related modules and analysis are given. Finally, in this paper's summarizes and prospects, the problems and areas to improve is analyzed.