说明: C:\Users\admin\Desktop\xwang8\index_files\clearpixel.gif

说明: C:\Users\admin\Desktop\xwang8\index_files\clearpixel.gif

Home

Biography

Research

Projects

Professional Service

Publications

Teaching

Students

Photos

Links

My Academic Genealogy

 

 

 

 

 

 

 

 

 

 

HiQuadLoc: An RSS-Based Indoor Localization System for High-Speed Quadrotors      

    Indoor localization for Unmanned Aerial Vehicles (UAVs) such as quadrotors has attracted much attention. However, the low-cost Wi-Fi RSS-based localization has not been applied on the quadrotors mainly because the high-speed flight reduces the number of times for RSS measurement. Moreover, the workload of collecting RSS training data is huge for 3-D indoor localization. In this study, we build HiQuadLoc, a new RSS-based indoor localization system for high-speed quadrotors.

    As shown in Fig. 1, our system consists of two parts: localization server and quadrotor client. The localization serveris connected to the Internet. A quadrotor which needs to useour system must have installed the quadrotor client and canconnect to the localization server through WLAN or cellularnetworks. Also, it has to be covered by at least two APs.During a flight, the quadrotor client keeps measuring RSS andother physical quantities periodically. The quadrotor client periodically uploads the newest metrical data to the localization server, and the server keeps estimating the newest location of the quadrotor. The latest localization results are replied to the client periodically.

Turning Detection

                
    To improve the performance of path estimation at the corners, the quadrotor client needs to detect the turning motion of the quadrotor. In our system we utilize the direction sensor installed on the quadrotor. Different from normal UAVs, it is harder for the client to detect the flight direction of quadrotors because a quadrotor can make lateral movements without changing its head direction. However, we notice that when a quadrotor is moving in a specific direction, its normal vector will have a drift angle for the same direction, as shown in Fig. 2(a). This is caused by the propulsion principle of the quadrotors. Thus we measure the normal vector instead of the head direction of the quadrotor.

4-D RSS Interpolation Algorithm in Offline Phase

               
    A typical RSS-based localization system needs to collect RSS training data during its setup phase. For normal 2-D based systems, the workload of training data collection is not critical, however for 3-D based systems, the workload increases with the height of a building besides its total area. Some technologies such as surface-based interpolation has been used to reduce the sampling rate of RSS for 2-D case, however most of them only record the average value of RSS at each calibration point, which loses the information included in the statistical features of RSS caused by the complex channel environment. Thus how to reduce the workload of 3-D RSS training data collection with consideration of the statistical properties of RSS is still an open question. In our system we propose a 4-D RSS interpolation algorithm, which can estimate the probability for a given RSS value to appear at a given cube.

Localization Algorithm

                                       
    The localization algorithm in our system consists of two parts: preliminary localization algorithm and path correction algorithm. The preliminary localization algorithm is based on the statistical properties of RSS fingerprints, and the 4-D RSS interpolation is applied to reduce efforts in the offline phase. Only low-precision results are provided by it. The path correction algorithm contains the methods of path estimation, path fitting and location prediction to further improve accuracy.

              

    The average location errors for these results are shown in Fig. 9(a). For normal localization systems based on RSS, their performance is the same with that of the preliminary localization algorithm. Thus if they are applied to the case of high-speed quadrotor directly, the average error will be 4.41m, which can hardly meet with the need of quadrotor localization. If we apply the path estimation method, the average error can be reduced to 3.16m. Moreover, the method of path fitting further reduces the error to 2.00m. Finally, considering the delay of transmission, the location prediction method achieves an average location error of 1:64m.

 


 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

说明: C:\Users\admin\Desktop\xwang8\index_files\clearpixel.gif

说明: C:\Users\admin\Desktop\xwang8\index_files\clearpixel.gif