Reseach And Design Of Mahjong Package Recognition System Based On Machine Vision

Mahjong packaging is the last process on the production line,however,currently the main product detection method,which relies on manpower as before,still has problems like large labour intensity and low speed.With the continuous development of automated industry,using machine vision instead of manual work for product defect detection has become the mainstream development direction of the industry detection field.Therefore,study of Mahjong package recognition system based on machine vision is an interesting topic with good innovation significance and practical value.In this thesis,templates entering and recognition algorithm of Mahjong package recognition system are first proposed.Then the hardware and software system framework are designed and realized according to the actual demand of the production line.Moreover,the stability and recognition accuracy of the system are quantitatively measured.The main work and results are summarized as follows:(1)The current research status on machine vision technology in industrial detection is deeply explored.Besides,analysis of the related product and research work on chess recognition is presented.Meanwhile,the significance of this thesis' work is summarized.(2)Two kernel algorithms in Mahjong package recognition system are introduced,which are templates entering and recognition algorithms.In terms of templates entering algorithm,by using the external feature of Mahjong,a whole rapid separation scheme based on the projection and a single object extraction algorithm based on shape feature are proposed.Furthermore,in order to construct the template library,an optimal template extraction scheme is presented by comparing the templates in same number.On the other hand,a coarse-to-fine pyramid SAD matching method is mentioned as well.Simultaneously,the correction methods of some misjudged special Mahjong are also analyzed.(3)The overall design plan of the hardware system is implemented on TI's TMS320DM6437 platform.Meanwhile,the primary hardware modules,including video capture and display parts,power supply and signal interaction designs,are analyzed and accomplished.System test results show that the stability and adaptability of the recognition system are enough to meet the detection quality on the packaging lines.(4)The overall scheme and process of the software system are designed.The main system function modules,including FLASH storage strategy,data exchange method,and OSD menu design,are researched and analyzed.In addition,the recognition algorithm is realized in the software system structure by C code,and the stability and recognition accuracy of the algorithm are quantitatively measured.Experimental results indicate that the accuracy of the identification achieves more than 98%.