Study On The Features-Based Method Of General Work-Piece Recognition And Its Key Technology

The technology of work-piece recognition has played an important role in almost all of the process of producing work-piece, such as reverse design, parts manufacture, assemblage and quality check, and has quiet wide applied future and research value. While using the technology of work-piece recognition, the efficiency of design and level of production automation can be improved, and the function of industry robot can be expanded to be more flexible. Moreover, for the multiformity of information of machine vision, the measurement system of automatic manufacture can be effectively reduced. Therefore, work-piece recognition has becoming one important pipeline of information in manufacture.But the actual work-piece recognition method is only applicable to some expert occasion, it is to say that if some new work-piece need to be recognized we must make out new recognition arithmetic using much time. For to resolve this problem, this paper propose a new method-general work-piece recognition based on future parameter formative, and selects it's some key technology to research, the following is the main work:(1) Research of feature class and class relizing method. Mostly work-piece can be composed of round,extrude,sweep and blend four feature, therefore this paper proposed based on future formative general work-piece recognition method, and analysed the feature's extract and choose which can differentiate that four feature, the neural net classifier is hereby designed, but some work-piece can't been differentiated correctly, so A novel method based on the neural network and principle component analysis(PCA) to generate feature set is present. In this method, firstly the neural network is used to generate new feature set utilizing the existed feature set, and then the dimension of new feature set is reduced by PCA, finally generalization capability criterion is used to judge the feature set's discriminabilital and easy dividual. Experiments volition show that the novel method can generate linear discriminabilital feature set, and generalization capability being remarkable, it can assure classification results being accurate.(2) The research of work piece picture's image processing. This paper propose to use mathmatics morphology to filter, this method not only can filter this noise such as gauss,salts and so on, but also can filter this noise who is evocable via the work piece oneself disfigurement to smooth work-piece edges and decrease false corner point; For to resolve this problem that the amount of corner points be extracted using harris method is redundant and its position is excursion, this paper propose an enhance method on the basis of harris method, it is replacing gauss filter operator by B-spline filter operator and replacing gauss differential operator by B-spline differential operator, experiment volition shows the corner points which is extracted by enhance harris method being reasonable and its position being exact; An improved edge detection algorithm on fuzzy sets is presented, it is take linear membership function and replacing enhancement function with gauss function to simplify the complex transformation calculation and solve iteration times r disfigurement. The experiment results prove that the algorithm is efficient,the acquisition edge is fine and false edges is very little.(3) This paper present work-piece stereo recognition arithmetic of each class feature based on binocular stereo vision on each class feature's three-dimensional model principle. And using a extend work-piece as a sample the paper present a detailed steps. The result of experiment shows the method being feasible and having small calculating quantity.(4) Research of stereo recognition method based on feature. To four features, according to the theory of 3-Dimensions parameters sculpt, the binocular based stereo recognition arithmetics of each feature are proposed. And the kind of extended work-piece is selected to research its arithmetic of binocular stereo recognition for the reason that the extended work-piece has much features elements that always present in another features. The proposed arithmetic is validated and the results show that the method is feasible and the account of calculation is little, especially the demands for the position of vidicon isn't strict.