Aplicación de redes neuronales de aproximación a una línea de luz para reconstrucción 3D de objetos

Abstract

A technique for 3D object shape detection based on light line image processing is presented. In this process, an approximation neural network is used to reconstruct the 3D object shape. This neural network is generated using images of a light line projected onto the objects, whose dimensions are known. These images are obtained in the scanning step of the light line onto the objects. The profilometric method used by the neural network is based on the light line deformations. These deformations are measured by the Gaussian approximation method. In this technique, the 3D shape is obtained without use the parameters of the experimental set-up. It is an advantage over conventional methods of the light line projection. In this manner, the accuracy is improved due to the errors are not introduces in the system. The accuracy in this technique is deduced by the rms value. This technique is tested with simulations and real objects. Also, the time processing and accuracy results are presented.