Design Of Switch Errors Prevented System Based On Machine Vision

With the extensive application of industrial and automation technology, technology ofmachine vision gradually emerges. Integration of automation and machine vision has openeda new page for the modern industry. The application of machine vision technology includesindustrial inspection, automated guidance, binocular vision and so on. In relation to humanvision machine vision has many advantages, lower cost, higher reliability, greater stability andmore. At present, industrial machine vision detection has become more and more attentive.In this paper, a system which can detect mistakes of the assembly switch was designed tomeet the requirements of a truck factory. Functions of the system include semi-automatedguidance and machine visual inspection. Semi-automated guidance guide workers to assemblethe switches according to the producing plans. Machine visual inspection system needs todetect whether the switches is installed correctly.Design of the overall system includes hardware and software. In the aspect of hardwarean industrial bench was designed and equipment was selected and assembled. The main workin software is to complete the detected software based on machine vision including therequirements analysis for both all the functions and non-functional requirements, selecting theappropriate image recognition algorithm and the design of the system working process, andfinally finishing software systems for tests and experimentation.Functional requirement is to detect weather the switches is assembled correctly,including the position of the switches installed and selection of the switches. However, in theenvironment of actual factory, the recognition results are susceptible to outside interference,which requires a strong anti-interference ability for the visual recognition software, so sixnon-functional requirements were list as reference for the design of recognition algorithm andsystem architecture.Visual inspection firstly detects the location of switches using a template matchingalgorithm. Template matching algorithm can return the coordinates of the area which is themost similar one to the template image in the image to be detected. During the process ofpositioning, the view of the camera is so large that it’s quite difficult to return the precisecoordinates of the switch. So the whole area which contains all the switches was locatedfirstly, then each switch is located in the area. The coordinates located by the templatematching algorithm should match the position information stored in the database. Gaussian smoothing, top-hat transform and thresholding are used to pre-process theimage. The appropriate methods are chosen by the algorithms. Top-hat transform based onmorphological is used to reduce the effects of the uneven illumination. The NMI algorithmand Hu moment invariants algorithm are firstly studied. Both of them are efficient fast andprecise in theory. Then SIFT algorithm is studied, SIFT algorithm recognition rate is high, butthe time complexity is slightly higher, and finally the PCA algorithm is studied, which cansimplify complex issues. PCA is used to reduce the dimension for SIFT operator to simplifycalculation. Then PCA-SIFT algorithm not only has similar recognition rate with SIFT butalso has a good time complexity. In order to verify which one is the most suitable algorithmfor practical applications,50pieces of images are chosen to have an experimental verificationafter comparison in theory. Thinking of the factors of accuracy and time complexity and so on,PCA-SIFT algorithm is actually the most suitable one for recognition.PCA-SIFT algorithm returns the most similar one with the test image in the templates.So the PCA-SIFT feature values of the templates are calculated and stored in the database.When the system starts, firstly downloads the production plan and then reads the informationaccording to the plan including the PCA-SIFT feature values, the signal range of each positioncoordinates and so on. After finishing assembling of the switches, the camera starts to capturethe image of switch panel. According to the plan it locates and recognizes one by one, andfinally returns the results.After the design of the system, hardware selection and software development are goingon. Then finish the installation of the industrial equipment and the debugging program of thesystem. And the result of an experiment of production and test is ideal. Finally summarize theworks and issues in the project and prospect the future.