Research On Key Technologies Of UAV SLAM In Smart City Under Dynamic Scenario

Under the current smart city scenario,the large-scale application of unmanned aerial vehicles in the modern urban environment relies on the robust positioning,detection and tracking modules.Precise positioning in the urban environment will be interfered by signal surges,building occlusion,feature repetition and other problems,which makes it very difficult.At the same time,traditional target detection and tracking methods also face challenges from dynamic environment,feature occlusion and distance estimation.In this paper,the algorithm of building map,locating,detecting and tracking modules of unmanned systems in urban environment is studied and discussed.In a complex outdoor environment,GPS positioning will be blocked by buildings,signal surges and other problems,resulting in errors in positioning accuracy.Due to the outdoor dynamic obstacles and the complex electromagnetic environment,the positioning stability and accuracy are highly required in the urban environment.Due to the limitation of the system structure,the simultaneous localization and mapping and the Inertial Navigation System are not robust enough.In this paper,a method is proposed to reconstruct the high-precision point cloud map from the large-scale urban road scene and realize accurate positioning by combining with the normal distribution transformation(NDT).The high-precision point cloud map can also be used in driverless cars and drones for sensing,positioning,autonomous navigation and smart city management systems.The application of high precision point cloud map can realize the location at the centimeter level in the city scene.This paper proposes a high precision point cloud map based on the surround lens for obstacle detection and completes the map acquisition and generation at normal speed on the expressway.With the development of deep learning,great breakthroughs have been made in the field of dynamic obstacle recognition.These technologies enable the dynamic point cloud obstacle detection and target tracking based on high precision maps to be effectively realized in the scene of modern smart city.It's Different from previous obstacle detection and tracking methods based only on vision,the use of high-precision map can greatly reduce the difficulty of point cloud classification,and the use of lidar can also solve the scale problem of dynamic detection.By combining the traditional kalman filter,Hungarian algorithm and deep learning,the robust dynamic obstacle detection and tracking is realized.Finally,the experiment proves that the system can effectively realize the problem of dynamic obstacle separation detection and dynamic target tracking in urban scene.This paper analyzes and discusses the principles and mathematical models involved in the above systems,introduces the data flow and key steps of the system,and demonstrates the experimental results to verify the principles.The results show that a complete and effective framework of unmanned information system in smart city can be realized by constructing high-precision point cloud map,NDT real-time matching and positioning,point cloud obstacle detection and tracking.It can effectively realize urban construction map,unmanned system positioning,automatic obstacle avoidance,object detection and tracking and other tasks.