| With the rapid development and application of unmanned aerial vehicle(UAV)and artificial intelligence technology,there are more and more researches on autonomous flight of UAV.Compared with the autonomous movement of ground robot in two-dimensional space,the autonomous flight of UAV in three-dimensional space is much more difficult,but its application prospect is also quite wide,such as search and rescue in complex terrain,monitoring of harmful substances and pollution sources,real-time monitoring of forest fire,highway and power line inspection,communication support,high-altitude fire fighting,express delivery,battlefield reconnaissance Aerial photography and agricultural plant protection.In the above application scenarios,due to the problems of UAV GPS signal loss,over the horizon flight,line of sight occlusion and communication interruption,it is impossible to use the remote control or external positioning system to control the flight.Therefore,it is very important to realize the autonomous flight of UAV.Therefore,based on binocular vision,this paper studies and implements an intelligent flight system which integrates autonomous positioning and obstacle avoidance of UAV(1)In order to complete the research of UAV autonomous flight,this paper designs and builds the UAV experimental platform,including hardware selection and software development.The d435i binocular camera on the platform is used for data acquisition of image and inertial measurement unit(IMU).The onboard computer processes the data and sends the control commands to the flight control in real time to realize the autonomous flight of UAV.(2)In order to realize the autonomous positioning of UAV,this paper selects the scheme of binocular vision and inertial navigation fusion for slam research.Firstly,the binocular camera and IMU are calibrated jointly to obtain their internal and external parameters and conversion matrix,so that their information can be better fused.The smaller the re projection error is,the better the calibration effect is.The re projection error in this paper is less than 0.5 pixels,which meets the use requirements.(3)In this paper,slam is implemented by the combination of front-end and back-end.Firstly,the front-end visual inertial navigation odometer calculates the pose of the camera through image feature extraction,feature matching and IMU data fusion.Then,based on the constraints,the back-end nonlinear optimization is used to obtain more accurate pose information.Finally,the bag of words model is used for Closed-loop Detection and optimization to complete the autonomous positioning of UAV.Through the test,the positioning accuracy is less than 10 cm.(4)Aiming at the current situation of semi autonomous flight which only studies autonomous positioning or uses external positioning system to study obstacle avoidance,this paper proposes an autonomous flight scheme which combines autonomous positioning and obstacle avoidance of UAV.The specific implementation process is to locate and plan the route for the UAV through the autonomous positioning system,and then use binocular vision to detect the obstacles in real time.When it approaches the obstacles,3dvfh obstacle avoidance algorithm is used to avoid flying around and finally reach the target point.Through experiments,the system realizes obstacle avoidance flight of UAV under autonomous positioning,achieves the goal of intelligent flight,expands its application scope and improves its survivability. |
