High-performance Segmentation In Moving Target Of Industrial Inspection Platform

Moving target segmentation is a technique that can be used to process images that are continuous in time,and the regions that be interested in the video object are extracted.It is the basic research in the field of pattern recognition.The image foreground target detection segmentation problem has always been in the model Recognition and computer vision occupy a pivotal position.A typical robot vision system in the industrial inspection platform.The moving target detection system is typically used to locate workpieces,components,tools,and other moving targets on the conveyor belt to capture workpieces,components,tools,and other moving targets,assembly,welding and packing.With the development of science and technology and the continuous improvement of industrial automation,the performance requirements of robot vision segmentation method on industrial detection platform are getting higher and higher,and the existed moving target segmentation algorithm,can not meet the requirements of accuracy and speed of industrial automation.In this paper,take the industrial robot visual inspection system as object of the research,and to get a high-performance segmentation method on the conveyor belt of the moving work piece target.The main contents of this paper include:Firstly,the basic theory of optical flow method and five typical optical flow field generation algorithms were studied,and also analyzed the advantages and disadvantages of them.The detection performance of LODF is qualitatively evaluated,and the optical flow method is chosen as the main method of motion detection and processing module.Secondly,the basic theory of super-pixel is studied,and the method of over-segmentation of dynamic image sequence is analyzed.The oversegmentation performance of SLIC is evaluated qualitatively and the SLIC algorithm is chosen as the main method to obtain the superpixel.Then,the algorithm flow and modeling method of the single Gaussian model and the Gaussian mixture model are compared and analyzed.The performance of the two modeling methods is evaluated qualitatively and the mixed Gaussian model is selected as the modeling method of the background model of the video frame.The accuracy of the algorithm is divided.Finally,this paper establishes a database of eight video sets,including the foreground targets such as unilateral workpieces,single rounds,double circular work pieces,double wrenches,and the establishment of its standard library for the database.The moving target segmentation experiment is carried out on the industrial platform robot vision detection system,and the segmentation speed and precision performance index of the segmentation method used in this paper are verified.In this paper,the method of fast object segmentation in unconstrained video is applied to the robot vision system of the industrial detection platform.By compared with some methods of robot vision system,like global threshold,Multi-scale-Normalized cut and grayscale symbiosis matrix,get the result that the method of this paper has a higher segmenta tion speed and accuracy,and achieved high-performance segmentation in moving target of industrial testing platform.