Research And Implementation Of Parts’ Feature Recognition And Classification’s Method Based On Machine Vision

With the development of modern industry, the parts’ sorting technology is developing in the direction of automation, high speed and network. However, the traditional artificial recognition and classification can’t meet the needs of modern manufacturing, and the sorting equipment of parts based machine vision can significantly improve the level of industrial automation and intelligence, which have many advantages, such as non-contact, high speed and high accuracy.This thesis focuses on the study and realization of parts’ features recognition and classification’s method based on machine vision. The key elements are as follows.1.The technology of parts’ preprocessing was studied. Some preprocessing operations which including image correction, image enhancement, smoothing, threshold segmentation, morphological operations, and edge detection were studied, and the results were analyzed combined with the part’s image in this paper.2.The methods of part’s feature extraction and classification were researched.After studying the common methods of image feature extraction, with the characteristics of parts, the feature extraction methods of shape and geometry features were mainly focused. Thereafter, the methods to identify and classify parts were studied, and the classifiers were designed using the template matching method and support vector machine method according to the shape to classify or according to the shape and size.3.The method to recognize the part’s pose was learned. A method to get the pose of part using minimum bounding rectangle of the part was proposed. After getting the pose in image, then it was converted to pose in actual world by camera calibration.4.The parts’ recognition and classification system based on machine vision was designed and implemented, including modules of parts’ image acquisition, image preprocessing, part recognition and classification, and pose recognition.5.It was tested to recognize and classify parts,and get the pose on the system.Experiments show that it has good versatility and computing speed using part’s shape and geometry features,and it’s feasible using the methods of template matching and SVM to recognize and classify, and it can meet the need of industry in positioning accuracy through part’s minimum bounding rectangle to recognize the pose.