Displacement equation-free laser triangulation ranging with deep learning

The measurement accuracy of laser triangulation ranging systems is easily affected by errors in system structure parameters, due to the asymmetric structure of the systems. This work proposes an intelligent laser triangulation ranging method that significantly reduces dependence on system parameters to increase measurement accuracy. Firstly, a nonlinear programming genetic algorithm with an elitist strategy is employed to design a laser triangulation ranging system. A deep neural network is then established for distance estimation without using the system structure parameters based displacement equations. Experimental results demonstrate this deep learning based laser triangulation ranging method can reduce the nonlinear error by the order of one magnitude, compared with the conventional method of using displacement equations. Meanwhile, the root mean squared error and the mean absolute error are reduced by factors of 5 and 3 to sub micron, respectively, which is no longer influenced by the design and setting errors of system structure parameters.