Research On Technology And System Of Vision-based UAV Autonomous Landing Navigation

Autonomous landing of UAV has very important meaning to its development and application,pose measurement and recognition of cooperative object are the core parts.In this paper,we research the technology and system of visually guided autonomous landing based on the practical demand of UAV's autonomous landing.First,addressing the problem that P3 P algorithm has multiple solutions and algorithm based on multiple points is not practical in engineering due to its working condition and performance,according to uniqueness of true value,we design a method of estimating pose by 2 group of P3 P algorithm based on 4 points.Due to the weakness of numerical stability,accuracy and efficiency in existing P3 P algorithms,SRP3 P algorithm with high computing efficiency is proposed by simplify the constraint equation in the foundation of SP3 P algorithm.And simulation results show that SRP3 P algorithm increases accuracy by 20% and the real-time by 39%.Secondly,starting from the engineering demand of The 4th International UAV innovation Grand Prix and in order to solve the problem of redundant computation and poor real-time,we propose a rapid resolution the yaw angle based on the rectangle object by combining perspective projection model and prix requirements.The result of the comparative experiment show that algorithm in this paper satisfy the demand of navigation precision and it's universal for all rectangle objects.Besides that it has higher efficiency contrast to traditional pose estimation algorithm and lower requirement in accuracy of image processing contrast to shadow point algorithm,thus the method has engineering application value.Then,addressing the problem in recognition of visible light cooperative objects under complex background,in the analysis of histograms of object image,according to the feature of multi-peak curves,we propose algorithm by selecting the bottom point between two peaks as the thresholds for crude extracts and then calculate connection area's geometric characteristics and textural features to extract the real objects.Experiments results show that choosing threshold value can be adaptive contrast to traditional segmentation algorithms and its real-time has been increased by 50% contrast to meanshift algorithm with otsu algorithm twice.In order to detect objects in all-weather,we select infrared lamps as cooperative objects,accordingly,we designed imaging system and detection algorithm to match,which have proven to be effective by experiments.Finally,addressing the lack of a complete system in present researches,we design and construct a vision-based navigation system for pose estimation to meet the demand.According to technical indexes and modular division by function,the implement of hardware and software has been designed in detail in this paper.And the test experiments show that the system can reach requirements of accuracy and real-time in the practical use of UAV landing navigation.