| Accurate environmental information is the basis for subsequent planning and control of intelligent vehicles.Existing SLAM（Simultaneous localization and Mapping）schemes often assume that external environment are totally static,but dynamic targets cannot be avoided in actual production and life scenes.In addition,most existing SLAM schemes are targeted at devices with sufficient excitation,such as drone,rather than intelligent vehicle platforms.All these factors may lead to the decline of the accuracy and robustness of intelligent vehicle’s localization and Mapping.In view of this,this paper studies a dynamic scene SLAM algorithm based on multi-sensor fusion for the intelligent vehicle platform.Through the fusion of three sensor signals of vision,wheel tachometer and inertial measurement unit,the synchronous positioning map is built.The specific research contents and results are as follows:（1）A multi-sensor fusion dynamic environment SLAM algorithm framework is constructed based on VINS-Mono,semantic information is introduced to identify potential dynamic targets,and wheel speed sensor is introduced to enhance the performance of the algorithm deployed on the intelligent vehicle platform.The algorithm reduces the influence of dynamic targets on system positioning through joint adjustment of multi-sensor information,and reduces the influence of dynamic targets on system drawing by eliminating dynamic features of visual information.（2）For image information,a feature extraction and homogenization algorithm based on adaptive threshold and adaptive regional grid size is proposed to obtain highquality features with uniform distribution and sufficient quantity.A dynamic feature elimination algorithm based on semantic and geometric constraints is proposed to eliminate dynamic features.（3）Preprocessing signals from different sensors.The IMU preintegration model and wheel speed preintegration model were established to establish the Motion constraints on the system states at different times,and the system pose and landmark position were estimated by the Structure From Motion（SFM）algorithm.The observability of the proposed algorithm under different motion excitation is discussed.It is shown that the wheel speedometer can not only provide odometer information for the system,but also make the system scale observable under constant acceleration,which proves the improvement of the proposed method compared with the existing schemes.（4）The system state parameters are estimated and the accumulated errors are eliminated.A robust initialization algorithm of vision-wheel speeding-IMU based on loose coupling was proposed.Based on this algorithm,the system state was optimized by joint optimization based on sliding window,and the accumulated errors were eliminated by closed loop detection and loopback correction algorithm.（5）The overall performance of the algorithm is verified by simulation and real vehicle experiments.The experimental results show that when the intelligent vehicle is taken as the carrier and there are dynamic targets in the surrounding environment,the positioning accuracy and robustness of the proposed SLAM algorithm are good enough,and the static three dimensional point cloud map can be established to exclude the interference of dynamic targets. |
