Research On Modeling And Motion Control Of Airport Runway Detection Robot

Simultaneously Localization And Mapping(SLAM)based on vision can provide unmanned systems with their current location information and observed environmental structure information,but it fails to make full use of the semantic information of the environment.Adding semantic information on the basis of VSLAM can not only increase the accuracy of the mapping,but also allow the unmanned system to better sense the surrounding environment,thereby completing more complex tasks.Therefore,this paper integrates the semantic segmentation information of 2D images into the 3D scene constructed by VSLAM to complete the construction of 3D semantic map.Firstly,the omnidirectional mobile robot platform is modeled by motion and the PID algorithm is used to achieve precise control of the robot.In order to meet the requirements for the free control of unmanned vehicles in the later mapping,the ROS system is used to realize the real-time of the PC to control the unmanned vehicles.Then,On this basis,the vision SLAM system based on Kinectv2 depth binocular camera is implemented,which mainly includes ORB?SLAM and RGBD?SLAM.Vision SLAM estimates the camera pose by matching feature points between two key frames,and uses the Bundle Ajustment algorithm to optimize the camera pose and feature point positions.In addition,a loop detection algorithm based on the bag of words model is used to determine whether the motion trajectory has a loop.If a loop is found,,The loop is added to the pose map used for global map optimization to eliminate errors.Finally,the optimized camera pose is obtained,and 3D map reconstruction is realized.Then a semantic segmentation system is designed to achieve the semantic segmentation of two-dimensional images.The PSPNet network is selected for semantic segmentation.The PSPNet network adds a pyramid pooling module on the basis of FCN,which makes up for the shortcomings of traditional semantic segmentation algorithms such as context mismatch and category confusion,and improves the accuracy of semantic segmentation.In this paper,resnet-100 is used for feature extraction of the image,and the resnet-50 in the original model is replaced to improve the segmentation accuracy.Finally,the environmental geometric information obtained by the visual SLAM system is fused with the two-dimensional semantic information obtained by the semantic segmentation system.The fusion method uses maximum fusion and Bayesian fusion to finally generate a three-dimensional semantic point cloud map.In order to save storage space and better express the map,the 3D semantic point cloud map is transformed into a 3D semantic octree map.With the movement of the Kinectv2 camera,a dynamic 3D semantic scene map is constructed.After the self-driving car completes the construction of the semantic map,it will have higher application value.On this basis,it can enrich the functions of the self-driving car,making the self-driving car more intelligent and humane.