Calibration of a Vision-Based Location system with Hybrid Genetic-Newton Method

To correct the uncertainty of the vision-based location system, a Hybrid Genetic-Newton Method (HGNM) is presented to calibrate its camera model. This method can minimize the uncertainty of the camera model by fusing the Genetic Algorithm (GA) and Newton method together. First, the camera model of the vision-based location system is built according to the image-forming rule and space geometry transformation principle of its visual measuring device. Second, the initial camera parameters generated by genetic process are iterated by Newton method until it meets the required accuracy. Otherwise, new populations will be generated again by GA and reiterated by Newton method. Third, a novel vision-based location system is designed to illustrate the application advantages of the modeling framework. The experimental result shows that the absolute error range of HGNM is [-1.1, 1.0] mm and the relative error range is [-9.49%, 0.11%]. It reveals that the accuracy of HGNM is about four times higher than LM method and up to six times higher than Newton method. In all, the HGNM is superior to traditional method when it comes to camera model calibration of the vision-based location system.