A Embedded Parts Detecting System Based On Machine Vision

In mechanical manufacturing field,it's a essential section to measure part's size.Fast and accurate detecting technology is the key to assure quality and efficiency.With the developing of modern industry,production scale expands gradually,and the requirement for parts is higher and higher,which presents new demand for detecting technology.Traditional tactile measurement needs too much time and power with low accuracy and artificial error,which meets demand of modern industry hardly for intelligence,informationization,automation,and high integration.Image vision technology is a hot research recently,which uses image sensor to collect image,then transform it to digital signal and adopts digital image processing technology to measure.Now,most image vision systems use PC and CCD image sensor to measure accurately,which is traditional with strong reliability,fast speed and high accuracy,and it occupied the field of image vision for long time.However,the image vision system has exposed its deficiency with extension,which limits its application with no modularity structure,poor transplantation,inconvenient installation,expecailly hard communication.The thesis proposes a embedded plan to measuer parts size based on image vision,which uses OV9650 image sensor to collect image information and invokes OpenCV to measure size in ARM.The researchs of thesis as follows:(1)Design harfware circuits based on the analysis of syatem's requirement and strucyure,including interface circuits of image sensor,memory and liquid crystal display,and other essential periphery circuits;(2)Transplant Linux operating system,including transplanting U-boot,Linux kernel and root filesystem.Design the start-up mode and analyse the driving principle of Nand Flash,LCD and image sensor;(3)Analyse the image processing,and transplant the OpenCV to process the image information which image sensor collects.Improve the way of image filtering with twice filtering technology-"Median Filter-Sharpening-GuassianFliter",then extract complete and clear edge information of circle on HCU;(4)Calibrate system with the method of calibration coefficient to transform pixel information to real size,then we gain the external diameter of circle.Finally,analyse the results with statistics.The experiment result shows that the standard deviation of measurement is 0.028mm,which can meet the demand of industry mechine vision size measurement fully.