Research On Multi-sensor Calibration For Unmanned System

In the current state-of-the-art solutions,the fusion of laser ranger finder(LRF)and camera is mainly used in the fields of unmanned driving and mobile robotics,and the calibration of the sensors is one of the essential parts.The calibration of LRF and camera is the first step to achieve sensor fusion,and the accuracy of calibration is a pre-requisite for successful sensing fusion.The calibration of LRF and camera is studied in this paper,and the main research contents are as follows.(1)Firstly,we present the current status of research on the extrinsic calibration of LRF and camera.Then we introduce the theory of two basic extrinsic calibration methods by comparing and analyzing these two methods based on the results of practical experiments.The experimental results show that the two methods have their own strength and weaknesses in respect to accuracy and complexity.(2)We design an extrinsic calibration method based on planar constraints.Based on the three-point registration method,the P3 P model is constructed.The calibration can be accomplished by using three sets of LRF and camera calibration data.The calibration results are evaluated from three aspects,including the use of the area constraints to find the true solution of the calibration results,the minimization of reprojection constraints,and boundary constraints based on the edges of the calibration plate.Then,by forming the equivalent constraint geometry,the invalid calibration input is reduced,which is a guideline for the calibration plate data acquisition in practical experiments.The simulation experiments show that this paper simplifies the calibration process and achieves higher accuracy by using boundary constraints.(3)We investigate the extrinsic calibration method based on enhanced calibration board.Then,we designed a tag for the calibration board.The extraction of the features for the camera and LRF is optimized.For the camera data processing,the calibration board parameters are extracted by the tag identification,and the calibration plate boundary is detected using the improved LSD algorithm.For the LRF data processing,the LRF scan point on the calibration plate is a straight line,and the straight line feature is extracted by threshold segmentation.Based on the working characteristics of LRF,the edge extraction strategy of the laser straight line is improved to get more accurate edge points by multi-frame laser data.The experiments show that the method in this paper has better robustness and practicality,and is more accurate compared with the existing state-of-art method.