Research On Visual Image Processing And Indoor Localization Technology

With the rapid development of computer and network technology,machine vision image processing technology has become a research hotspot in the field of artificial intelligence and pattern recognition.In recent years,visual image processing technology has been widely used in indoor localization,automatic driving,artificial intelligence and other fields,so the research of this technology has important significance for the development of modern science and technology.In this thesis,the image dimension reduction technology,image retrieval technology and indoor vision based localization technology are developed.The main work and innovations are as follows:1.There is a lot of redundancy in the high dimensional raw images,which not only greatly increases the computational burden of image classification process,but also inevitably degrades the classification performance of the model.To solve this problem,a novel feature selection model is explored for dimension reduction.Firstly,a structured sparse penalty term is added to the LRSSR model to reduce the influence of image noise.Then,LRSSR solves the rank minimization problem through Schatten-p norm,so the model can learn the inherent low-rank structure information in the original data,select representative features and reduce redundant features.The results show that the proposed LRSSR model can effectively select the robust low dimensional features,and can obtain the high classification accuracy,which lays a good foundation for improving the accuracy of classification retrieval in the process of image retrieval algorithm.2.Traditional retrieval technology based on image matching is inefficient.Although the image retrieval technology based on neural network has fast retrieval speed,it is greatly affected by light,scale and rotation changes,and the retrieval results are not ideal in indoor scenes.To solve this problem,a fast image retrieval(FIM)method based on image classification is proposed.Firstly,a hybrid classification model is proposed to improve the accuracy of image classification.It can reduce the scope of image retrieval.Then,SURF algorithm is used for image matching.It can solve the problem of traditional retrieval method based on neural network and can improve the matching accuracy.The results show that the improved image retrieval method has better accuracy and efficiency,which lays a good foundation for vision image-based indoor positioning.3.A lot of noise exists in indoor machine vision images,which not only greatly degrades the accuracy of image feature matching,but also inevitably increases the estimation error of fundamental matrix.To solve this problem,an improved vision image-based indoor positioning(IBIL)method is proposed in this paper.Firstly,an outlier matching point elimination algorithm(OMPE)is proposed to eliminate outlier matching point pairs based on correlation coefficient threshold constraint.It can eliminate the negative influence of outlier on fundamental matrix estimation.Then,an improved fundamental matrix estimation algorithm(RFME)is also proposed.It can restrain the fundamental matrix by algebraic error and geometric error so as to reduce the estimation error of the fundamental matrix.Finally,a machine vision image-based indoor localization method(IBIL)is combined with the above new algorithms.The results show that the proposed indoor localization method can effectively improve the localization accuracy.