Abstract: For the three-dimensional surface scanning robotic system, the system measuring accuracy mainly depends on the following aspects: Robot positioning accuracy, line structured light vision sensor measuring accuracy, hand to eye calibration accuracy. The three aspects mentioned above can be considered as part ones. Whatever how much error will these three aspects be, the final error will be reflected in the obtained measuring data. However, the error model of the system is difficult to be established via mathematical models. As to enhance the system measuring accuracy, the error model is constructed via particle swarm optimization-radial basis function neural network(PSO-RBFNN). The input of PSO-RBFNN is chosen as the image coordinates of laser stripes, while the output of PSO-RBFNN is chosen as the errors between measured points and closest points after iterative closest point technique. The centers and widths of RBFNN are optimized by particle swarm optimization(PSO) to find more suitable parameters and thus reducing the number of neurons in the hidden layer. Experimental results demonstrate that the system measuring accuracy is improved after the error compensation scheme via constructed error network model.

Key words: Error modeling and compensation, PSO-RBFNN, Robot, Three-dimensional surface scanning