Research On SLAM Method Of Rescue Robot Based On Multi-source Information Fusion

Complex environments such as natural disasters and battlefield rescue have increased the difficulty for ground rescue personnel.Traditional rescue methods are difficult to ensure the safety of rescuers and timely and accurate completion of rescue tasks.Rescue robot is a necessary choice to carry out rescue tasks in complex environment instead of human,so it is extremely urgent to develop a SLAM system that can help rescue robot achieve robust autonomous localization in complex scenes.At present,visual SLAM technology is developing rapidly,but there are still some shortcomings in the complex rescue environment.For example,in the case of fast movement,image interruption and motion blur,visual SLAM system will lose posture tracking.The visual SLAM system does not work properly during rapid rotation and drastic light changes.The autonomous localization of rescue robot in complex environment cannot be achieved by relying on a single sensor.The multi-sensor fusion SLAM technology has become the main research direction of autonomous localization of rescue robot at present.Therefore,a multi-source information fusion SLAM system based on camera,IMU and wheel speedometer is designed in this paper,which enables the rescue robot to have robust localization and mapping ability in complex rescue environment.This paper carries out research from the following aspects.Firstly,in order to solve the tracking loss problem of visual SLAM system in special environment,a fast relocation algorithm based on ORB-SLAM is proposed.When the robot has motion blur,image interrupt and fast movement,SLAM system is easy to lose tracking.At this time,it is necessary to correlate the feature information of visual image with the constructed map information to realize the system relocation.An "Image to Map" relocation algorithm was designed based on the framework of ORB-SLAM.The experimental results show that the improved ORB-SLAM relocation algorithm has better robustness and real-time performance than the original ORB-SLAM relocation algorithm,and the relocation efficiency is improved by about 8%.Secondly,a visual inertial SLAM system based on ORB-SLAM is implemented with direct method and common view optimization.The visual SLAM system does not work well when the camera rotates rapidly,the light changes frequently,or even when there is no light for a short time.IMU can measure high frequency motion information,and the camera and IMU have good complementary characteristics.Based on this,a tightly coupled visual-inertial SLAM system is implemented.The initialization of visual inertial SLAM system and tightly coupled pose estimation algorithm are studied.Experimental data show that in terms of real-time performance,the proposed algorithm does not rely on descriptor matching between feature points and does not need descriptor calculation.Compared with ORB SLAM3,the initialization is faster,but it has a slight disadvantage compared with VINS-Mono.In the pose estimation phase,the proposed algorithm is 50% faster than the other two algorithms,and 26% faster than the Vins-Mono algorithm based on optical flow tracking.Finally,a multi-source information fusion SLAM system based on vision /IMU/ wheel tachometer is designed by incorporating the wheel tachometer into the visual inertial SLAM system mentioned above.Visual inertial SLAM system is affected by zero acceleration bias and gravitational acceleration,and observability degradation occurs in linear and non-rotating motion for a long time,which reduces the positioning accuracy of the rescue robot,while the wheel tachometer can accurately measure the speed and yaw information of the rescue robot in linear motion.Therefore,the integration of wheel tachometer on the basis of visual inertial SLAM system can effectively improve the localization accuracy of SLAM system.Experimental results show that the multi-source information fusion SLAM system designed and implemented in this paper can effectively improve the autonomous localization ability of rescue robots in complex scenes,compared with pure vision and visual inertial SLAM systems.