Research On SLAM And Cooperative SLAM Based On Visual Inertia

With the continuous development of science and technology,Simultaneous Localization and Mapping(SLAM)technology has become a hot topic.At present,the single robot SLAM algorithm has made a lot of progress,and has been gradually applied to production and life.However,there are still some problems need to solve,poor dynamic stability and so on.Further,due to the complexity and enlargement of the mission scene,single robot SLAM often fails to meet the task requirements.CSLAM share information with each other,which greatly improves information utilization and enhances environmental awareness,which has great advantages in a wide range of unknown environments.For this reason,this paper studies the technology of SLAM based on visual inertia and cooperative SLAM,and designs experiments to verify the effectiveness of the algorithm.The specific research contents are as follows:Aiming at the problem of poor accuracy and time consuming in traditional visual odometer,a visual odometer integrating optical flow and feature matching is proposed.Firstly,the traditional reference frame/current frame tracking method is prone to generate accumulated errors,a local optimization algorithm is introduced to estimate the camera's initial pose.Secondly,a unified loss function of optical flow / feature points is constructed based on the key frame to optimize the camera's pose.The experimental results on the Eu Ro C dataset show that the accuracy is equivalent to that of the feature point method in simple environment,and the accuracy is higher than that of the feature point method in case of missing feature points.The test results of running time show that the algorithm can reduce37.9% of the running time compared with the feature point method on the basis of ensuring the accuracy.Aiming at the problem that visual odometer is easy to fail under the condition of fast movement and sharp change of illumination,the visual inertia tight coupling method is introduced.Firstly,IMU is modeled,kinematics formula is constructed,and camera data and IMU data are matched by pre-integration method.Secondly,the IMU data constraint equation is constructed in the sliding window,and the joint optimization function is formed by combining vision and IMU residuals.The experimental results on the Eu Ro C dataset show that the proposed method can stabilize the input pose and reduce the overall positioning error.According to the characteristics of multi robot system and the requirements of scene mapping,a multi robot cooperative mapping algorithm is proposed.Firstly,a two-level road sign fusion and update algorithm is proposed with the help of Closed-loop Detection ideas.Secondly,combined with the position and attitude estimated by the positioning system and the depth information obtained by sensors,the environment is constructed with three-dimensional dense map.Experiments are carried out in simulation and real scenes.The results show that the algorithm can improve the location accuracy of road signs and effectively construct 3D dense map.In order to evaluate the performance of positioning and cooperative mapping algorithm in the actual scene,a set of robot hardware experiment system is built,and each sensor is calibrated;from the perspective of positioning accuracy and mapping effect,three groups of real scene experiments are designed from simple to complex.Compared with the orbslam2 algorithm,the end-to-end error of this algorithm is smaller,and it can build the 3D dense map of the surrounding scene clearly and completely.