Technology Research Of Sealed Tank Leaking Detection Based On Machine Vision

In China, there are many seal vessel manufacturers. an important means to detectwhether a qualified products or not is to detect the leakage of seal vessel. Thetraditional approach is to put the seal vessel into water manually and observe thebubble to determine the leakage of seal vessel. But manual inspection has large laborintensity and low work efficiency. According to the insufficient of traditionaldetection method, this paper gets small seal vessel as an object and targets themovement of air bubbles in water. then study the methods of moving target detectionand recognition based on Image processing theory. And discuss methods applicationsin practice. Completed work of this paper is as follows:1. Research at home and abroad, summarizes the characteristics of the structureand movement of the target of a small airtight seal vessel. To solve the problem oflarge estimation error of background value when the lack of training for classicGaussian mixture model and high target misjudgment, this paper proposes the methodof background value initialization based on K-means clustering theory. take meanclustering of the sequence of pixels in images as the initial value of backgroundevaluation image, determine the quantity of Gaussian distribution which describeimage pixels change based on the number of clusters. Then initialize the parameters ofthe Gaussian distribution. The experimental results verify the validity and accuracy ofimproved Gaussian mixture model applied in bubble motion region detection.2. According to slow iterative, computational complexity and calculation error inoptical flow calculation method, this paper proposes improved Horn-Schunck opticalflow calculation method based on brightness constancy constraint and localsmoothness constraint. Construct a selection function. According to the magnitude ofthe gradient, apply different assumptions constraint conditions. Introduce localsmoothness constraint which reduces the optical flow calculating error. Experimentalresults show that improved Horn-Schunck optical flow calculation method reduces theamount of optical flow calculation and improves accuracy in recognition of bubblemotion area. 3. apply improved Gaussian mixture model target detection algorithm andimproved Horn-Schunck optical flow calculation method. get moving target regionfrom image segmentation based on background model. get the movement trend ofmoving region from Optical flow calculation. corrected direction of the optical flowfield according to the movement of characteristics of the bubbles, then identify thebubble region. get the position and number of bubbles by Statistics in the outline ofthe image based on Freeman chain code method and leak coordinates of the sealvessel. Finally, achieve leak bubble detection, recognition and localization. Set up theexperimental platform and carry on experimental verification: This method caneffectively detect the leak state of seal vessel and provides theoretical support fordesign and implementing of automatic leak detection system.