Banknote Recognition System Based On Embedded Platform

Along with the social progress and development,more and more ma-chineries have realized automation,intelligentization and informatization.In-telligent currency recognition system can improve work efficiency significantly by implementing paper currency discrimination,defect detection,currency val-ue identification and serial number recognition as a substitute for labour.There-fore,portable,low power consumption and cost-effective embedded recogni-tion system is getting more and more popular in medium enterprises.This thesis engages in a in-depth research on embedded currency recognition sys-tem.In the first,according to the image properties,an identification algorithm based on HOG feature and support vector machine is proposed.This algorithm can achieve a high recognition precision.To meet the computational speed needs,another identification algorithm based on OHV and template matching method is proposed,it can achieve a recognition precision of 99.9%.Secondly,according to the platform features of DM642,an algorithm transplantation is accomplished.In order to improve computational speed,multi-level optimization for this system is accomplished,including C code op-timization and DSP platform specific optimization.As a result the work,the processing time per frame of value recognition drops from 500 ms to 6 ms after highly optimization.Thirdly,this thesis made a study of usd dollar,euro serial number recog-nition.A series of work are accomplished,including image preprocessing,serial number vertical location,characters segmentation,correction,and recog-nition.The experimental results show that a recognition precision of 99.5%is achieved,indicate the algorithm is robust and highly adaptable.