Research On Classification And Recognition Of Quad-rotor UAV Landing Sign Based On Vision

With the development of science and technology,autonomous landing technology of UAV(Unmanned Aerial Vehicle)plays an important role in various fields,and the classification and recognition technology of landing sign is the premise of UAV autonomous landing.In order to improve the UAV’s ability to classify and recognize the landing sign,this paper further discussed the image classification model based on convolutional neural network(CNN: Convolutional Neural Network).On premise of this,it focused on studying the scheme that let the classification of the landing sign first and then the recognition in the manner of feature purification,thus accurately extracting the characteristic corner point coordinates of the landing sign and making the classification and recognition of the UAV all the more accurate and effective.This paper mainly does the following work:First of all,analyzed the principle of quadrotor UAV,and designed a distinctive landing sign.Aiming at the limitations of conventionally only using image recognition technology to recognize landing sign,a scheme is proposed to first classify the landing sign image,then perform the feature corner extraction,and finally perform the classification and recognition of the pose estimation.To improve the accuracy and speed of landing sign classification and recognitionSecondly,aiming at the shortcomings of traditional image processing technology that cannot accurately determine the landing area,constructed a landing sign image classification model using convolutional neural network(CNN),exclude non-landing sign areas.And introduced a multi-layer structure on the basis of the existing image classification model,based on the image classification model of different layers,introduced the batch data volume and the learning rate μ.Discussed and analyzed the impact on model training speed and accuracy.After experiments,the model’s image classification accuracy rate reached 98.18%.Then,aiming at the problem of landing sign identification,Applied SURF algorithm based on Harris corner detection information.Aiming at the problem of precision and speed of feature extraction,use the RANSAC algorithm to purify the features extracted by the SURF algorithm.The feature extraction speed is increased by 32%,and the accuracy is increased by 23%.At last,for the problem of pose parameter estimation,introduced the coordinate system of the quadrotor UAV.For the accuracy of feature corner coordinates,the method of sub-pixel positioning is applied.Aiming at the problem of system implementation,Using Open CV library,C++,QT,Python to design the visual software,and conducted real-time and accuracy tests,the results show that the pose estimation error of the visual system designed in this paper is 0.4cm,the average running time is 1.17 s.The accuracy and speed of classification and recognition of landing sign are improved.