Research On Self-localization Algorithm Based On Monocular Vision Motion Carrier

With the improvement of the degree of social information,people have more and more demand for location information.For positioning systems in outdoor scenes,GPS,Beidou and other technologies have better positioning in outdoor environments,but not suitable for indoor scenes,and vision-based positioning technology has received wide attention because of the relatively simple equipment required and the small impact of environmental changes.In the research of vision-based indoor positioning technology,image matching is a key part.The speed,accuracy and robustness of indoor positioning are directly affected by the speed,accuracy and robustness of image matching.Therefore,this paper mainly conducts in-depth research on image matching technology and indoor positioning technology based on monocular vision.Firstly,the subject has been calibrated for the camera.In the research of self-positioning technology based on the motion carrier of monocular vision,calibration of the camera is a very important step.The accuracy of the visual positioning result is affected by the accuracy of the camera calibration and the robustness of the calibration algorithm.Therefore,the basis for the accuracy of the results based on visual positioning technology is to make the calibration of the camera better.In this thesis,the linear calibration method,the nonlinear calibration method and the two-step method are compared.It is concluded that the method of Zhang Zhengyou calibration in the two-step method can calibrate the camera only by using the checkerboard at different angles.The calibration method is simple and the precision is high.Therefore,the camera is calibrated by the method of Zhang Zhengyou calibration in this thesis,and the internal and external parameters of the camera are obtained.Secondly,this topic improves the AKAZE algorithm based on the SURF algorithm.In the self-localization process of motion vector based on monocular vision,image matching plays an important role in the process of visual positioning.SIFT,SURF and KAZE image matching algorithms use Euclidean distance for descriptor matching,and the matching accuracy is high,but real-time is poor,ORB,BRISK,FREAK and AKAZE algorithm uses Hamming distance for descriptor matching,the matching speed is fast,but the accuracy is relatively poor.In this thesis,the feature point detection in the Hessian matrix of SURF algorithm is combined with the local differential binary descriptor(M-LDB)in AKAZE algorithm,and the image matching is realized by Hamming distance.The matching results in this paper are more robust than SURF and AKAZE algorithms in image matching with different degrees of blur,different JPEG compression and different viewing angles.At the same time,the matching time under different image transformations is also shorter compared with the matching time of SURF and AKAZE.Finally,the subject achieves self-localization based on monocular visual motion vectors.The information collection of the indoor environment is carried out,and the environment feature database is constructed,and the combination of the global feature descriptor and the local feature descriptor is used to achieve the purpose of image search.After obtaining similar images,the polar geometry algorithm is used to realize the positioning of the motion carrier,and the improved AKAZE,AKAZE and SURF image matching algorithms are used to compare the positioning time and positioning accuracy of the positioning.It is concluded that the algorithm is better in terms of positioning speed and accuracy.So it is suitable for positioning systems.