Study On Calibration Algorithms On Robotic Multi-Sensors Based On Swarm Intelligence

With the wide applications of robots in the domains such as autonomous driving,aerospace,medical services,and exploration and survey,it is hard to satisfy the work requirements in complex unstructured scenes for robots only with single or single-type sensor.To accurately perceive the scene,and ensure the sufficiency and accuracy of robot for scene understanding by using multi-source and heterogeneous sensors,has become an important issue in the field robot study in recent years.Sensor calibration is the basis for obtaining accurate data and multi-sensor information fusion.However,existing calibration methods are greatly affected by measurement noise and outliers,and the calibration accuracy and consistency need to be improved.Therefore,deeply study on the multi-sensor calibration of robots has important theoretical significances and practical values for improving the accurate perception and understanding of the environment for robots.In this thesis,the problems of three-dimensional(3D)laser scanner calibration,and the self-calibration of extrinsic parameters between camera and 3D laser scanner are deeply studied.The calibration parameters are solved by using the swarm intelligent optimization algorithms such as the invasive weed optimization(IWO)and the success-history based parameter adaptation for differential evolution(SHADE).The main study contents are summarized as follows:(1)A calibration method of 3D laser scanner based on pseudo Huber loss function is presented,which can estimate the measurement model parameters of the 3D laser scanner.Based on the analysis of the measurement model of 3D laser scanner,the space sphere is used as the calibration object.And the objective function is established according to two factors,that is,the constraint that the distance between the spherical surface point and the spherical center is equal to the radius,and the pseudo Huber function is insensitive to the outliers.Then,the combination algorithm of IWO and Levenberg-Marquardt(LM)is adopted to optimize the objective function.Experimental results demonstrate that the proposed method can effectively suppress the influence of measurement noise and outliers on the calibration results.(2)A calibration method of 3D laser scanner based on SHADE algorithm is presented.Three strategies,that is,local filtering weighted plane fitting(LFWPF),K-nearest neighbor(KNN)clustering and threshold segmentation are adopted to effectively remove most outliers in the calibration data.Then,the SHADE algorithm with excellent performance is adopted to optimize the solution by using the calibration data with outlier removement.Experimental results demonstrate that the proposed method can effectively remove outliers,minimize the influence of measurement noise and outliers,and improve the accuracy and consistency of model parameter estimation.(3)A SHADE algorithm-based self-calibration method of extrinsic parameters between camera and 3D laser scanner is presented.According to the correspondence of the visible laser scanning lines both in two-dimensional(2D)image and 3D point cloud,an optimization objective function without special calibration object is established.And the SHADE algorithm is adopted to solve extrinsic parameters between camera and 3D laser scanner.Experimental results demonstrate that the proposed method can yield accurate extrinsic parameters.