Small Fixed-wing UAV Vision Navigation Algorithm Research

With the rapid development of UAVs,their application scenarios are not only confined to the military field,but also expanded to civilian fields such as power inspection,agricultural forestry plant protection,logistics and transportation.The above processes often have requirements such as long endurance time and heavy load,so small fixed-wing UAVs can be better applied to the above scenarios.On the other hand,small fixed-wing UAVs also have disadvantages such as poor flexibility,high control difficulty,dangerous take-off and landing.Therefore,it is particularly important to design a safe and effective navigation algorithm.Most of the existing navigation algorithms are for indoor multi-rotor UAVs,while small fixed-wing UAVs have high lighting instability in outdoor environments.What’s more,fixed-wing UAVs fly very fast.Therefore,this paper hopes to design a fast navigation system with light robustness.Based on the above research background,this paper designs navigation algorithms for small fixed-wing UAV cruise process and landing process respectively.The details are as follows:1)Optimize the tightly coupled S-MSCKF algorithm based on extended Kalman filter for the characteristics of small fixed-wing UAV cruise process.Considering the lighting instability of the outdoor environment,a robust feature point method that is not affected by lighting is introduced to implement the front-end feature matching process.On the other hand,in order to compensate for the decrease in detection efficiency caused by the feature point method,in the back-end optimization part,an effective frame discrimination algorithm is introduced to remove the redundant update process to achieve performance optimization.Finally,use the UPeen dataset for offline flight verification.2)Design a detection and tracking algorithm of the target warehouse.The UAV123 dataset is augmented with data augmentation.Building A with hemispherical white arc tops and building B containing green football fields are selected as two types of target warehouses.Then,use YOLOv3 target detection method to achieve warehouse detection.On this basis,perform pruing with the channel slimming method on the YOLOv3 network to improve the real-time detection effect.Finally,in view of the problem of target occlusion that may occur during the cruise,a Deep Sort target tracking algorithm is introduced to perform detection and tracking experiments with pruned YOLOv3 network.3)Propose a vision-aided small fixed-wing UAV autonomous landing algorithm.Runway detection is implemented using probabilistic Hough transform and redundant line segment screening.On this basis,a runway geometric model is established,and the straight line equation of the runway center line in the pixel coordinate system is obtained.In order to ensure that the runway is still within the camera’s field of view during the rolling process,introduce the geometric constraints on the rolling angle.Furthermore,a dual-sensor fusion mode is designed to use valid visual information and correct system errors in GPS,so as to ensure that landing navigation can also be achieved when visual information is invalid.After determining the position of the runway relative to the UAV,a state space model of the small fixed-wing UAV is established and the subsystem parameters are identified.Based on this,the trajectory planning algorithm and the attitude control algorithm based on predictive control are designed to achieve attitude tracking.Finally,select the Jinshan UAV test flight center sfor actual flight verification.