The recent widespread interest in robotics originates from the motivations: to increase productivity, reduce cost, overcome skilled labor shortages, provide flexibility in batch operations, improve product quality and to free humans from bornings, repetitive tasks or operations in hostile environments. At recent, however, most of the jobs performed by robots are simple, repeatable, monotonous tasks. If we are to apply them for more complex tasks, it will be necessary to add many more functions, especially intelligent functions. Among the several ways to implement the intelligent robot system, robot system, robot vision fields are mainly adapted.
In this thesis, recognition of printed music score with large amount of music data and its practical application to robot vision was studied.
The system consists of multiple steps: music score image enhancement, stave position detection, thresholding, stave removing, coarse classification, fine classification, music syntax check, and interface to robot. Firstly, in preprocessing the music score image, stave detecting in the gray level image and adaptive thresholding method are proposed. Secondly, in coarse classification of music symbols, the integral projection on the horizontal line are used. Thirdly, in fine classification and music syntax check, tree structured algorithm for rapid recognition time and auto modification method for correcting music syntax error are proposed. Last, in interfacing to robot for performing the music score by robot, fast code generating and finger position determining algorithms have been developed.
As a result of this study, it was shown that the music score with 100-120 symbols per one frame is feasible for automatic reading and the recognition accuracy is about 95-98 percentages and recognition speed for performing by robot is satisfactory.