Design Of UAV Autonomous Navigation System Based On Semantic VSLAM

With the development of high technology,people have higher and higher requirements for robots,such as autonomous fetching and autonomous walking.Traditional indoor positioning methods,such as Ultra Wide Band(UWB),Bluetooth,WIFI,etc.,can only provide simple location information.Robots do not understand objects in the environment and cannot navigate autonomously more intelligently.Traditional UAV systems cannot quickly and accurately conduct autonomous navigation in indoor low-light complex environments.In this regard,this article focuses on the research and design of an efficient flight control navigation system based on visual semantic maps.The main work and innovations of this paper are as follows:1.The accuracy and robustness of the traditional Vision Simultaneous Localization And Mapping(VSLAM)system in low-light environments will be seriously affected.In response to the above problems,this paper combines the LIME algorithm to enhance the input image,so that the VSLAM system can reduce the maximum drift of 1.1m in the low-light environment,improve the accuracy of 0.12 m,and increase the robustness.2.The UAV uses an embedded platform,which has certain restrictions on computing power,memory resources,and power consumption.Generally,the low-complexity neural network model leads to insufficient accuracy.In response to the above problems,this paper combines data enhancement,transfer learning,initial learning rate setting and cosine learning rate adjustment strategies on the basis of lightweight neural network Yolov4-Tiny.After training and optimization,the average accuracy of the network model has increased by 6.6%.3.When constructing a semantic map,traditional 3D point cloud recognition is directly mapped to a 3D point cloud through a 2D tag,which causes the recognition of objects to be blurred.In response to the above problems,the system maps the two-dimensional object annotations to the three-dimensional dense point cloud after supervoxel pre-segmentation,and realizes the three-dimensional annotation to form a semantic map with higher efficiency,and the time cost of the comparison algorithm is increased by 1.7 times.And 3.2 time4.The trajectory planned by the traditional path planning algorithm is difficult to adapt in a complex environment,and it is easy to fall into a local solution,resulting in unsatisfactory trajectory planning or planning failure,and the planning speed is slow.In view of the above problems,this paper uses local planning algorithm combined with global planning algorithm and Fast-Planner to optimize the trajectory,so that the planned trajectory is more in line with the flight control flight,and compared with the traditional path planning algorithm,the time cost is based on Fast.-Planner’s planning method has been improved by 1.8 times.Finally,an efficient autonomous navigation system based on semantic maps is realized on the Pixhawk hardware control platform,PX4,and ROS software control platform.The main functions of the system include object recognition,semantic map construction,autonomous navigation and so on.This article has done a lot of experiments to verify the accuracy and effectiveness of the system.