Shape mensuration and recognition by DDS approach

A systematic approach for shape/profile mensuration and recognition is presented in this paper. Object shape and profile, represented by data sampled from its boundary, are coded into autoregressive (AR) models and then analyzed by the data dependent systems (DDS) methodology. The geometric features of a shape/profile, such as sharp corners, are extracted together with the frequency domain characteristics. Periodic feature of a profile can be measured in the image plane and converted to the world plane by means of the optical mensuration. New descriptors, composed of the characteristic roots, are used for recognition purposes. The results show that the descriptors together with the feature-weighted algorithm significantly improve robustness of recognition which is essential in a typical noisy environment on the shop floor. The approach has been implemented on an applicable PC-based package with currently available optical devices and computers.