The Research Of Embedded Machine Vision System Based On Image Features

The machine vision which based on the computer is widely used in the visualinspection applications, it is familiar with its speed, accuracy and stability. But thesmall and medium-sized enterprises in China cannot afford the traditional machinevision system and do not have the ability to develop on it. In order to combine theadvantages of machine vision and embedded system, we devote ourselves to developa cheap but powerful embedded machine vision development platform that suit for thesmall and medium-sized enterprises.According to the requirements of image matching, the common image algorithmis designed standardized, parametric and realized in the PC and embedded systemprogramming respectively, for completing basic image algorithm library. On thefoundation of the study about image contour, skeleton, feature point detection andmatching features, an image local invariant feature detection method, which is basedon genetic algorithm and an image invariant feature description method that aims atobtaining the number of feature points, values, gradient as the main feature vectorsare proposed, along with a Features fast image matching algorithm which is based ondeclining feature weights is proposed. Besides, the CASS embedded machine visiondevelopment platform is completed and it can meet the needs of different industrialvision projects. While, the embedded vision module can calculate its matching degreewith template image without checking the characterization description of thematching image.It is regarded as the advantages of using the features fast imagematching algorithm which are proposed by this paper. As result, the vision modulecan be used alone.The development of the checking project of part gesture in factory assembly lineuses the CASS embedded machine vision development platform. The rapiddevelopment process proves the usefulness of the development platform.97%of thereal-time correct part gestures recognition probability shows the superiority of theimage matching algorithm proposed in this paper.