Computer vision approach to three-dimensional curved surface measurement and surface representation for object recognition

The field of artificial intelligence examines the capabilities of computers in the performance of intelligent tasks. The solution to many problems in artificial intelligence may be obtained through the observation of a human performing the task in question. The adaptation of sensory equipment to computers adds further capabilities to simulate human perception and task performance. One powerful computer peripheral which has only recently been made cost effective and adaptable is the vision system. With computer vision systems, the capabilities of computer perception are greatly enhanced, and its potential for the performance of intelligent tasks is more apparent.

The application of computer vision systems to the measurement of three-dimensional objects is a powerful tool for such applications as robotics, navigation and surveillance. The basic concepts of stereo vision have long been understood for measuring polygonal solids. How-ever, certain complications arise when an attempt is made to apply these techniques to the measurement of curved surfaces. The primary focus of this thesis is the presentation of various techniques for curved surface measurement and representation. Stereo vision system principles and their applications to curved surface measurement through "active imaging" are discussed. Techniques for obtaining surface shape from shading information are also described. Finally, a discussion of methods for representation and recognition of curved surfaces is given. The significance of this study is the presentation of surface measurement and object recognition techniques which are directly applicable to system automation in the military and in industry.