Multi-Sensor Calibration For Mobile Vehicles

For mobile vehicles,multi-sensor information fusion is a key technology to ensure their longterm intelligent operation in complex and changing environments,the essence of which is the multilevel and multi-space information complementation and optimal combination processing of multiple sensors with different modalities to finally produce robust and accurate state estimation.One of the prerequisites and necessities is the highly accurate extrinsic calibration of multi-sensors,which is the rigid transformation matrix between two sensors,based on which the multi-sensor observations in the system can be converted to the same reference coordinate system.The calibration between the external sensors has the requirement of common field of view between sensors,while the trajectory alignment calibration method without the requirement of co-visualization is difficult to achieve high accuracy calibration.The calibration between the external sensors and the body sensors requires full motion excitation of the body sensors on the one hand,while on the other hand,the external sensors tend to lose the field of view under full motion,and the trade-off between these two presents a challenge to the calibration work.In this thesis,we investigate the calibration problem between the above multi-sensors for commonly used external sensor(visual camera,Li DAR),and body sensor(IMU),and the following innovations are achieved.(1)To address the problem that the existing camera-Li DAR sensor extrinsic parameters calibration requires the common field of view between sensors,while the trajectory alignment method is difficult to complete high precision calibration,the spatio-temporal decoupled Li DAR-camera calibration under arbitrary configurations is proposed.By matching the all-view information of the Li DAR and the panoramic view generated by the camera movement,more number of calibration targets are observed to produce more data association,which effectively improves the accuracy of the extrinsic parameters calibration.At the same time,the proposed method lifts the installation restriction of sensor field of view overlap,because the change of sensing unit position extends the camera field of view,and the Li DAR sensor can still have common observation with the extended camera field of view even if there is no co-view between sensors at the initial position.In addition,this thesis designs the calibration process to decouple the time delay component and eliminate the effect of time asynchrony between the two sensors.On the basis of this,this thesis derives the observability of the calibration system and gives the optimal calibration plate placement scheme,which theoretically guides the calibration work.The validation results on simulation tests show that the proposed calibration method has a translation error less than 0.01 m and a rotation error less than 1°,which is better than other compared methods.(2)To address the problem that the target area in the long-focus camera image is small,the corner point detection error is large,and the telemetry Li DAR observation data is sparse,which makes it difficult to complete accurate matching between the two to complete the calibration,this thesis proposes an ultra-long range narrow-view camera-Li DAR calibration method.By constraining the detected pose of the calibration plate as an optimization variable,and in order to avoid inaccurate point-to-point data association introducing erroneous information,the proposed method only needs to construct more robust plane-to-plane data association for constraint under long-range observation.After project validation,the proposed method improves the precise extrinsic parameters calibration method to serve the metro advanced driver assistance system,realizes the calibration of ultra-long-range long-focus narrow-view angle,solves the problem of accurate correspondence of visual small targets on Li DAR points,and is important for the advance long-range obstacle detection of fast-moving unmanned metro monitoring system,which provides technical support for the safe operation of mobile vehicles.(3)A multi-camera-assisted high-precision camera-IMU calibration method is proposed to address the problem that the existing camera-IMU extrinsic parameters calibration method requires full IMU excitation,and the visual camera tends to lose the calibration plate field of view when the IMU is fully excited,resulting in inaccurate estimation of motion and difficulty in achieving highprecision calibration.By introducing auxiliary cameras into the camera-IMU system for assistance to form a multi-camera-IMU system,the full range of environmental information is collected and more environmental observations and constraints are introduced.Although the additional information provided by the auxiliary cameras can improve the estimation accuracy,it also brings more variables to be estimated and thus affects the calibration accuracy.In order to investigate the impact of introducing auxiliary cameras on the calibration,this thesis theoretically demonstrates that using more cameras in the calibration can achieve better calibration accuracy,using the lower bound of the covariance of the extrinsic parameters to be estimated as a measure.The simulation test results show that the calibration accuracy of the proposed method can reach millimeter level,and the error is reduced by 70% compared with the existing calibration methods.When the IMU and the camera are configured at a farther distance,the error can be reduced by more than 90%.(4)Aiming at the intelligent and autonomous operation of mobile vehicles with multi-sensor fusion,this thesis takes rail transit subway and unmanned detection of special mobile vehicles as research objects,builds a multi-sensor hardware platform,deploys a multi-sensor sensing system,and realizes the verification of the calibration algorithm of the above research.Among them,for the existing unmanned subway state estimation methods are based on 6-degree-of-freedom motion models or planar motion models,which lack the modeling method of fast-moving orbital models,the curvature constraint-based real-time odometer estimation of unmanned subway is proposed.Orbital constraints are applied to the motion model by embedding the curvature constraints to the orbital environment into the constraint terms of the odometer estimation.The more common line features in the subway environment are also introduced for assistance,which greatly reduces the drift and improves the estimate accuracy.Aiming at the sensing requirements of unmanned detection of special mobile vehicles,this thesis completes the sensor selection and arrangement of multi-Li DAR,multi-camera and IMU,and forms a set of multi-sensor information fusion state estimation system demonstration.After testing,the system design scheme is correct,the hardware selection is reasonable,the configuration is complete,the technical path is feasible,and the functional index requirements of the project are satisfied.In summary,this thesis presents the key technology of calibration and information fusion between Li DAR-camera-IMU for the application demand of intelligent adaptation of mobile vehicles to achieve autonomous operation,realizes the millimeter-level calibration of multiple sensors,tests and passes the acceptance of the proposed algorithm in several projects such as rail transit subway and unmanned detection of special mobile vehicles.It provides theoretical exploration and experimental conclusion for fusing the information of multiple sensors to realize the long-term autonomous operation of mobile vehicles in complex and changeable environment.