In manufacturing industries, machine vision is introduced in assembly and inspection automation to lessen human intervention and to increase system flexibility and productivity. An important feature of assembly automation is real-time capability. Most industrial tasks require high speed processing capabilities. Therefore, it is required to use high speed processing hardware and fast and efficient algorithm.
In this thesis, methods of recognizing two-dimensional objects are proposed by using the concept of interrelation quadruplet. The interrelation quadruplet is invariant under translation, rotation, and scaling of a pair of line segments. The geometrical relationship of a simple m-vertex polygon can be represented as m - 1 interrelation quadruplets. And also, several useful properties of the interrelation quadruplet are derived in relation to fast and efficient object recognition.
A simple occlusion-free objects are recognized fast and efficient manner by using the concept of the interrlation quadruplet. But, recognition method requires a polygonal approximatiun stage to use the interrelation quadruplet. For recognizing simple object, it is required to recognize more simple and efficient way. For this, a method of recognizing simple occlusion-free objects is proposed by using the concept of generalized incremental circle transform. The generalized incremental circle transform, which maps the boundary of an object into a circular disc, efficiently represents the shape of the simple objects. And a vector function is derived which is invariant under object translation, rotation, and scaling using the generalized incremental circle transform. This function is used as feature information for recognizing two-dimensional simple objects efficiently.
Recognition of shapes which are incomplete or distorted is important in many machine vision applocations. Multiple objects, touching objects, or overlapping objects can cause the partial occlusions and occlusion causes significant problems in identifying and locating the objects. In this thesis, a recognition method of occluded objects by using the several properties of interrelation quadruplet is also proposed.
Object families have moving subparts. In this case, the same object may have different shape depending on the status of the moving suparts. In this thesis, several properties for recognizing object families are introduced and recognition method of object families using the interrelation quadruplet are also proposed.
The proposed methods require only small space of storage and are shown to be computationally fast and efficient.