Research On Recognition Of Workpiece And Sketch

Workpiece recognition is to automatically classify workpieces on the production line with computer vision technologies. This technique can help industrial robot to achieve tasks with high efficiency and accuracy, which will further reduce the cost of production and boosting the industrial production. Such automatic operation includes workpiece sorting, setting up, transporting, treating, etc. The workpiece recognition could be abstracted into sketch recognition problem. Sketch is a special category of images, which is the sparsest form of all images. Generally, sketch classification has a great deal of important applications, especially for human-computer interaction, children education and law enforcement. The existing sketch recognition methods mostly regard sketch classification as a sub-class of pattern recognition, and apply the methods for general images on the sketch classification problem. Although the methods did work, they are not good enough to be applied in real life applications.Workpiece recognition and sketch recognition methods are studied in the thesis, and the main work is as follows:(1) Combining with the application situation of workpiece sorting and setting up on the production line, a new workpiece recognition system is proposed. In this scheme, the algorithms for image pre-processing, segmentation, feature extraction, and pattern recognition are exploited. Besides, a corresponding image pre-processing method is presented with full consideration of the influence of the environment, such as the illumination, which provides a foundation for the workpiece recognition system to effectively segment workpieces and extract the contour information. An integrated experimental and testing platform for workpiece recognition is built and the corresponding software system also is implemented, which achieve good performance and satisfy the real-time processing requirement.(2) A sketch recognition method based on deep learning is proposed for the application of sketch. This method makes full consideration for the special characteristics of sketch(sparse, color, etc.), and builds a new system with convolutional neural network. The test results on the YU-Berblin dataset show that the proposed method has a good performance for sketch recognition.