360-deg profile noncontact measurement using a neural network

A new approach to automatic 3-D shape measurement is presented and verified by experiments. This approach, based on neural network theory, can automatically and accurately obtain the profile of diffuse 3-D objects by using a projected laser stripe. When the laser stripe is projected on an object, the line image of the laser light is grasped by a CCD camera. Using neural network theory, a relationship between the laser stripe image in the CCD camera and the related absolute position in space can be established. Thus the spatial coordinates of a measured line image in a CCD camera can be obtained according to the output value of the neural network. By processing a series of laser line images from the discrete angular positions of an object, a complete 3-D profile can be reconstructed. Theoretical analysis and experimental systems are presented. Experimental results show that this approach can determine the 360-deg profile of an object with an accuracy of 0.4 mm.