Research On Low Cost Visual Detection Platform And Key Technologies Of Visual Detection

With the development of technology and the decline of hardware cost,the visual detection is applied more and more widely.Surface defects detection and workpieces assembly orientation detection are important to ensure product quality.Compared to the high cost,low efficiency and false detection due to visual fatigue of artificial detection methods,visual detection technology can achieve detection automation,improve production efficiency.Considering the large-scale application needs to control the production cost,so the choice of visual detection platform should take into account the low-cost,maintainability and processing speed.Therefore,combined with surface defects and workpiece orientation detection,this paper studies the low-cost visual detection platform and the key technologies of visual detection.Firstly,the key algorithms of image preprocessing are studied in this paper,including image matching,image enhancement,image segmentation and so on.For image matching,the template matching based on cross-correlation and the image matching based on feature points are studied respectively,and the image angle correction is implemented by using feature points matching.For image enhancement,the adaptive histogram equalization algorithm to improve the contrast of the image,the homomorphic filter algorithm to eliminate the uneven illumination of the image,and the color constant algorithm to reduce the image color cast are given.For image segmentation,the image segmentation based on HSV color space and the OTSU segmentation algorithm based on gray space are studied.By calculating the H,S,and V components corresponding to each color in the standard color card to get the threshold of the three-channel components of the common color,and dividing the region of interest corresponding to the specified color.Combined with OTSU algorithm,the recursive OTSU and the block OTSU are studied,which makes the OTSU algorithm can achieve better segmentation results when the background is complicated.Secondly,the surface defects are complex and difficult to detect.It is observed that defects often exist in regions with obvious local changes in the uniform background of the image,which is in line with human visual attention mechanism.So combined with the image texture and visual saliency model,a detection algorithm of surface defects based on texture saliency is proposed.The algorithm using the image block as the basic processing unit,extracts GLCM and GGCM features based on statistical texture.The eigenvalues are classified by K-means clustering algorithm,using Gmean as a performance evaluation standard,and the texture features with high contribution rate are screened out to calculate saliency map,merging into the final saliency map and extracting the defects.Compared with the existing four kinds of saliency detection methods,the algorithm can detect strong defects and weak defects at the same time,and effectively suppress the interference of background noise.The precision ratio,the recall ratio and overall F-measure of the proposed detection are 0.751,0.856 and 0.794,respectively.The surface defects in the image are extracted accurately and effectively.Then,on the choice of low-cost development platform,through the comparison of the cost,performance and commonality of the commonly used development platforms,Raspberry Pi based on Cortex-A53 architecture is selected as the system hardware platform and Raspbian based on Linux kernel as the operating system.Taking the orientation of the workpiece in the actual production line as testing object,this paper designs the hardware,software and processing algorithm of visual detection.In hardware respect,including camera selection and workpiece imaging program design,light source selection and lighting program design.In software respect,adopting QT and OpenCV to design the overall system.In algorithm respect,based on the analysis of the characteristics of the workpiece image,in view of the characteristics of small size and no significant difference between the two ends,using the light to make the workpiece gray-scale differences at both ends,designing an algorithm for judging the obverse and reverse surface based on the proportion of binary pixels in the bevel after workpiece segmentation and an algorithm for judging the direction based on the variance of gray values at both ends of workpiece.Finally,the workpieces in different states are tested to verify the feasibility and efficiency of the visual detection system for workpiece orientation.At the same time,it also shows that Raspberry Pi has the advantages of low cost,high universality and high integration as visual detection platform.Finally,in view of the universality in image acquisition and image processing of field applications,in order to enable visual detection system developers to focus on the specific test object algorithm research,based on IPC processing platform to build a visual system development environment,developed a universal visual detection system.The system integrates camera calibration,image acquisition,common preprocessing algorithm,surface defects detection and workpiece orientation discrimination algorithm in one,supports multi-camera working at the same time.It can do most of the basic work of visual detection,and can be redeveloped for the specific test objects.