Strategy Of Multi-sensor Information Fusion Navigation In Complex Scenes

With the continuous development of artificial intelligence technology,mobile robot research has received more and more attention,many research results can perform well in specific scenarios.However,when facing the task under the complex and diverse working conditions,the requirements for the stability and accuracy of the autonomous navigation of mobile robots are higher.The existing solutions of individual sensor are often limited by data sources.There are many problems in practical applications.Therefore,this paper mainly studies the fusion strategy of multi-source heterogeneous sensors to realize autonomous navigation of mobile robots in complex environments.According to the measurement model of different individual sensors,after completing the construction of the hardware platform,the external parameter relationship between the coordinate system of each sensor was calibrated.The data is described and processed in a unified coordinate system,and the time axis is also calibrated.Through experiments,we have completed the positioning performance analysis of four sensor modules: IMU,encoder,lidar,and satellite positioning.In order to sovle the problem of positioning in GPS restricted environment,based on satellites' number detection and UTM coordinate system conversion,a seamless connection process was designed.Aiming at the unstructured environment,based on the feature of undisturbed positioning results derived from the encoder track,and combined with the PLICP odometer,a detection method of difference comparison and threshold judgment is proposed.We used encoder's track deduction odometer to replace laser positioning in unstructured environment.The strategy makes the positioning performance more robust.At the same time,based on the aforementioned two research contents,a multi-sensor loosely coupled SLAM scheme based on EKF and a multi-sensor tightly coupled SLAM scheme based on factor graph optimization were respectively proposed.The loose coupling scheme occupies less computing resources but is mainly used for plane motion.The tightly coupled scheme can cope with the six-degree-of-freedom movement on the uneven ground.Finally,our schemes were verified by physical experiments,which shows the effectiveness of the proposed strategy,the accuracy and robustness of mobile robot positioning and mapping were improved.Method of path planning based on A* algorithm and DWA algorithm was designed,so that the mobile robot can autonomously navigate,track,and avoid obstacles on the basis of SLAM.For mobile robot navigation in highly dynamic and interference scenarios,the aforementioned SLAM algorithms have poor performance.A motion strategy combining target person detection and visual following was proposed,so that the mobile robot can follow the person to move out of the area.In order to reduce the demand for computing resources,a fast target person detection algorithm based on Haar-like features,Cascades cascade classifier,and nonmaximum suppression was designed.At the same time,the control algorithms of the PTZ and the mobile robot were designed.For situations where mobile robots cannot be manipulated on site,a teleoperation platform has been developed to realize remote control,and we proposed a way to compress and upload the occupied grid map.Finally the effectiveness of the algorithms was verified through experiments.