Research On Key Technology Of Stereo Visual Perception And Autonomous Navigation For Micro Air Vehicles

With the rapid development of sensor technology,communication technology,artificial intelligence technology and other high-tech,micro aircraft have become more autonomous and intelligent,and have been widely used in military and civilian fields.In order to ensure the safety and reliability of micro aircraft in complex flight environments,it is necessary for micro aircraft to sense the surrounding flight environment and autonomous navigation in real time so that micro aircraft can complete tasks such as obstacle avoidance and path planning.Therefore,according to the volume and weight limitation of micro aircraft,this paper selects the optical sensor,and research the key technology of the three-dimensional environment sensing method based on binocular camera,and the three-dimensional environment sensing method based on single camera and binocular camera.At the same time,based on the visual environment perception and inertial sensor information,the visual odometer is constructed to output the pose and position information of the micro-miniature aircraft stably,which provides information for the aircraft to achieve reliable control.In order to make the micro aircraft use the binocular sensor to sense the three-dimensional environment effectively,based on the analysis of the optical camera imaging model,the fixed baseline parallel binocular stereo vision depth perception model is studied.Aiming at the stereo matching problem in binocular perception,the framework of local stereo matching algorithm is analyzed.The disparity credibility calculation is added into the disparity optimization algorithm.On the basis of ADCensus matching cost and adaptive matching window generation,a local stereo matching algorithm based on disparity credibility is designed,the accuracy of binocular visual 3D environment perception is improved.The binocular sensing algorithm enables the micro aircraft to have the ability to calculate the distance between the aircraft and the flight environment in real time,while providing the data foundation for visual odometer pose estimation.According to the analysis of the error model of depth estimation for the parallel binocular sensor with fixed baseline,the long-distance environmental perception result is not credible because its depth error is too large.Based on the analysis of the polar geometry constraint between two frames,the triangulation model is established.The image feature extraction and feature point matching method are studied.The feature point mismatch culling algorithm with distance threshold and RANSAC algorithm filtering is studied,while the visual pose estimation method is also researched.Without other sensors,using the 3D environment information perceived by the binocular camera,combined with the continuous sequence image acquired by the left camera,a three-dimensional environment-aware method based on single/binocular is proposed,so that the binocular environment sensing algorithm breaks through the limitations of fixed baselines and increases the range of perception.This approach allows the aircraft to sense the flight environment effectively.The pose estimation information of micro aircraft in real-time is the basis for its autonomous positioning and navigation.Under the complex flight conditions such as satellite rejection,it is difficult to satisfy the navigation demand,which is micro aircraft estimate their position and pose using traditional inertial and satellite navigation methods.By analyzing the principle of visual odometer and its recursive calculation characteristics,the pose estimation method based on joint nonlinear optimization is designed to improve the accuracy of visual odometer pose estimation.In view of the instability of the visual odometer due to the large maneuvering movement of the micro-small aircraft or the sudden change of the flight environment,the inertial-assisted visual odometer and the environmental prediction sensing method are designed,so that the visual odometer can provide continuous pose estimation information for the micro aircraft.This approach enhances the robustness of visual odometers and the safety of micro aircraft.The paper uses a small six-rotor aircraft,equipped with ZED binocular vision sensor,manifold processor from DJI company,and other hardware equipment to build a micro aircraft visual perception and autonomous navigation method verification platform.The 3D world image is taken as the experimental data by the binocular camera.The binocular 3D environment sensing algorithm and the single/binocular 3D environment sensing algorithm are used to make experiment and the sensing result is analyzed;By collecting the visual odometer pose estimation results of micro aircraft in real environment,as well as the real-time data of inertial sensors,differential GPS and other sensors to analyze the trajectory of the aircraft.The experimental results and data verify the effectiveness of the proposed algorithm.