Navigation And Localization Method Based On Heterogeneous Sensor Information Fusion

Due to its advantages of small size,high speed,good concealment,and strong survivability,UAV are increasingly used in military and civilian fields.Navigation technology of UAV is the basis for safe flight and is a key component of UAV systems.UAV’s navigation technology mainly includes localization,obstacle avoidance,object detection and path planning.This paper focuses on the localization algorithm and object detection of UAV.In indoor environments,UAV cannot use localization methods such as GPS.The choice of sensors has limitations due to the payload and power of UAV.Inertia-based Inertial navigation system(INS)and vision-based simultaneous localization and mapping(SLAM)have their own disadvantages.The composite localization method which combine the INS and SLAM system has become the main method of indoor localization.This paper designs a UAV indoor loosely-coupled localization system,which depends on Kernel Adaptive Filtering method to integrate inertial sensor data and vision sensor data,to improve the indoor localization accuracy of UAV.Fast and accurate detection of the object’s species and position in the image are prerequisites for improving the efficiency of the UAV’s grasping and transporting.Accurate detection of the optimal capture frame for different objects is the guarantee to improve the success rate of the UAV’s grasping.The traditional UAV grasping research frequently uses external devices to acquire position and attitude information between the UAV and the object,such as using a motion capture system,but these methods have their own limitations.This paper designs an object detection system for UAV grasping.The system uses the object detection algorithm based on convolutional neural network to provide the species and position information of object for UAV,and controls the UAV to fly over the object.The RGB-D sensor is used to acquire the object’s image and depth information,and an optimal grasping box is trained based on the deep learning network to provide the grasping information for the UAV.This paper discusses and analyzes the principles and mathematical models involved in the above systems,introduces the system’s frame and key steps.We verifie the effectiveness through experiments,analyze and discuss the experimental results.The results show that the indoor localization system can improve the localization accuracy of the UAV,and the object detection system can provide the species,localization and optimal grasping box of the object.