This paper primarily presents a new approach to the quality control inspection of 3-D objects by means of 2-D camera views. The approach permits a simplified computational method for shape matching. The inspection system utilizes this method of data acquisition first to store its model image of the “perfect” object and later to acquire data from inspected objects. Inspected objects whose shapes do not meet similarity specifications after matching with the model are rejected.

The system was tested using 17 objects, each of which was inspected 3 times, for a total of 51 trials. The system produced correct classifications in 45 trials, so the reported rate of success was 88.2 percent. From the experiment, it appears that the approach achieves a great speedup of the inspection process. This is made possible by 3-D object inspection using simple 2-D imaging, shape boundary extraction, and shape matching techniques without involving any complicated 3-D data acquisition, 3-D representation transformation, or 3-D shape matching. Having reduced the dimension from 3-D to 2-D, it becomes possible to see the generalized Hough transform [1] for shape registration and distance-weighted correlation [2] for similarity measurement. The hardware and computational facility requirements are also very simple, and further improvement in the success rate appears to be technically feasible without great expenditure.

However, there are also some disadvantages. Although the small number of trials reported (51) is enough to establish that the system functions, it is probably insufficient to establish the reliability of correct classification results. More trials with more objects may produce a significantly different (probably better) result. Further, the authors acknowledge that one limitation of the system is that it only compares boundaries or silhouettes of the object and does not inspect surfaces, so some types of defects will not be discovered.

This is a rather interesting paper that contributes to both pattern recognition and artificial intelligence. It should attract readers with research background in these two fields. Perhaps, in the future, if such research and development can involve knowledge-based rules for industrial parts inspection, it will make some more significant contributions in robotics and knowledge engineering [3] and even in the factory automation of the 1990s [4,5].