Research On Feature Extraction And Key Frame Retrieval Of Video GIS Data

With the rapid development of smart cities and the advance of city safety and defense demands,the question that how to accurately discover and retrieve the data from VideoGIS data faces a series of bottle-neck problems.Therefore,the analysis and utilization of these data are important.At present,content-based videoGIS retrieval often uses a single feature and a two-feature matching method,which is often unsatisfactory in the face of massive videoGIS data.Based on videoGIS data,for the problem of repeated and redundant video frames in videoGIS,this paper extracts valuable videoGIS key frames and videoGIS features,and then builds key frame library to reduce the amount of videoGIS index data.At the same time,global features,local features and deep features are extracted respectively to construct a videoGIS feature library.Finally,we efficiently realize the key frame retrieval and get the retrieval results with the high quality so that the potential value of videoGIS big data could be fully utilized.In this paper,the main research results are as follows:(1)VideoGIS hierarchical organization model and structured processingAccording to the characteristics of videoGIS data and the requirements of efficient retrieval,a hierarchical organization model of videoGIS data is designed.The videoGIS data is divided into three levels: video,shot and key frame.Aiming at the problem of repeated and redundant video GIS frames,a videoGIS shot boundary detection algorithm for simultaneous detection of shot abrupt transition and gradual transition is proposed.And then,the videoGIS key frame extraction algorithm is designed by using the Euclidean distance frame difference.(2)VideoGIS data retrieval based on multi-feature fusionA multi-feature fusion retrieval method is adopted.The feature extraction of videoGIS keyframes includes global features and local features.In addition,the feature vectors are constructed by combining global features and local features as multi-feature fusion representations,and then similarity ordering is performed.By adopting the method,the correlation between videoGIS features is fully utilized.And the method has a good retrieval effect.(3)VideoGIS data retrieval based on deep learningThe deep neural network is used to extract the videoGIS deep feature,and obtained the features are coded by using hash function.The proposed retrieval method is implemented by adopting the way from coarse leval to fine level,so the retrieval speed is improved under the condition of ensuring accuracy.The experimental results show that the algorithm can retrieve the target quickly and accurately.