Research On Simultaneous Localization And Mapping Technology Of Indoor Mobile Robot

Due to the aging of the population and the aggravation of labor shortage,robots are developing an emerging field.They are not only widely used in the industrial field,but also gradually entered into people's life.More and more robots need to perform autonomous work in random and uncertain scenarios with the continuous development of the society,and the technical requirements in this aspect are also increasing.In this context,SLAM has gradually become the core technology for indoor mobile robots to work in unknown scenarios.At present,most SLAM systems use laser radar or vision sensor to estimate the pose of robots,but both of them have their own advantages and disadvantages,so this paper studies SLAM algorithm based on those two kinds of sensors.Firstly,This paper derived the mathematical basis of spatial data conversion for robots.Then calibrates internal parameter of the camera and the external parameter between the camera and laser radar sensors.And the LM method is used to optimize the calibration results.And the robot model and measurement system were built on the TF tree model which could realize the rapid transformation of spatial data.Secondly,SLAM system model based on two-dimensional laser radar is established by describing the robot's motion in the control equation and observation equation.The bayesian probabilistic model is used to solve the problem of positioning.And a batch processing algorithm for particle positioning is proposed,which saves the computational resources for optimal position and pose matching.And this paper completed mapping on incremental grid construction method by establishing relationship between preprocessing index and the occupied probability value of grid.The grid occupation status is updated by searching the index value,which improves the construction speed of grid map.After that,the vision-based SLAM framework was divided into front-end and backend processing.In the front end,the method of random sampling consistency is used to eliminate false matching points and improve the matching accuracy of image features.EPnP is used to estimate camera frame motion.In the process between two frames pose estimation,a local map is established to increase the correlation of those tow frames' data.In the back end,the global optimization strategy based on position and pose graph is used to reduce the computational burden of the system and make the global position and pose estimation more accurate.The loop detection based on word bag model is added to the system,which effectively reduces the cumulative error of the system.Finally,an experiment was designed in the actual environment to test the proposed lidar based SLAM algorithm and visual sensor-based SLAM algorithm.The validity of the two algorithms was verified by location experiment and mapping experiment,and the experimental results were analyzed.