Research On Indoor Positioning Of UAV Based On Visual SLAM Algorithm

Robots are moving from the laboratory to our life.Unmanned aerial vehicle(UAV)is the most representative one.It is widely used in meteorological survey,highway inspection,exploration and mapping,flood monitoring and so on.The positioning of UAV without GPS signals becomes an urgent problem.Because of its low cost,simple equipment and strong adaptability,SLAM has become a better choice.So far,There are many SLAM algorithm have been published.But the algorithms in the 3D space,using camera remains to be studied.At present,there are fewer visual SLAM systems that can build maps and locate robots in real time under the condition of using high performance processors.Based on the existing research,a novel algorithm is proposed.The algorithm is equipped with the Linux development platform,using xtion Pro,a RGBD camera produced by ASUS as the visual sensor,and the depth error is calibrated.The algorithm based on optimization theory is taken as the core algorithm.The front end uses Tracking algorithm based on ORB feature.When building the map,IMU is used to fuse into the system.Loop closing in the back end use bag of words based ORB feature.Finally,the construction of unknown space maps and robot localization is achieved.This algorithm has the following advantages over traditional algorithms:1.With the improved camera calibration algorithm,the accuracy of the spatial point position of the corresponding pixel is greatly improved.2.By fusing the information of IMU into visual SLAM,we can provide relatively accurate initial values for the iteration and improve the localization accuracy of the robot.3.The initialization part abandons the traditional method by using fundamental matrix and use a relatively reliable method by five point algorithm.4.Advanced programming techniques,such as multi-thread parallel programming and SIMD programming have been used to improve computing performanceThe algorithm can acquire images,locate the robot and construct the map in real time only by UAV and a RGBD camera on it.Compared with other localization algorithms,it has the advantages of simplicity and reliability.Experimental results show that the positioning error of this method is less than 5% within a short time,and it can meet the requirements of positioning.