Research On The Detection Method Of Panoramic Street Image Change

The study of panoramic street image has become a hot issue in the field of remote sensing.Change detection of panoramic block images can investigate violations and illegal constructions of urban streets scientifically and accurately,and also make a scientific and reasonable plan for the transformation of shantytowns and old urban areas,which has the great value to the planning of the whole city.In this paper,the panoramic block image before and after the tsunami earthquake obtained by the vehicle camera is used as the experimental data.According to the summary of the existing change detection methods,three methods are selected to study the panoramic Street image through preprocessing,change information extraction,accuracy evaluation and other processes.The main work,research contents and results are as follows:(1)Aiming at the current situation of less research on panoramic block image change detection at home and abroad,the methods of panoramic block image change detection based on pixel,panoramic block image change detection based on object and panoramic block image change detection based on depth learning are studied,and carry out relevant experiments.(2)Research on pixel-based change vector analysis(CVA)detection of panoramic street image changes: according to the theoretical basis and mathematical principle of the change vector analysis method,,after the change vector analysis method,the Otsu method is selected as the threshold segmentation method to determine the change threshold.Through this method,the better threshold segmentation results are obtained.Finally,the accuracy of the two groups of experiments are 60.3% and 58%,respectively.(3)Investigate the change detection method of panoramic block image based on Support Vector Machines(SVM): use post-classification comparison method to detect the change of panoramic block image,the method of selecting the support vector machine in the experiment,selecting the training samples according to the type of feature,and supervising the classification.The accuracy of the experiment is 72.8% and 78.9% respectively.(4)Research on the detection results of panoramic street image changes based on the Segnet networkmodel: aiming at the problems of traditional change detection methods,such as low accuracy in panoramic block image detection and poor quality of the obtained change result image,this paper adopt the change detection method based on deep learning.The main selection is segnet network model,which is suitable for the data in this paper.The specific method is to classify the training set into 11 categories,and through training and testing,the accuracy of the two sets of experiments is 81.4% and82.2%,respectively.The experimental results show that: among the methods selected in this paper,the change detection accuracy based on deep learning is improved compared with the other two methods,but it still has not achieved the desired effect.It can be seen that there is still a huge challenge to the change detection of panoramic street image.