Abstract: In order to efficiently organize and manage the increasing real-time sensor data and associations, and satisfy the requirements of multi-level tasks for multi-dimensional feature calculation and association mining of multi-modal scene data, a spatiotemporal sparse graph index method is proposed for the bottleneck problems of disk I/O-intensive, low processing efficiency and weak support for associations existing in the tree structure based external indexing methods. Firstly, a spatiotemporal index structure based on in-memory graph model is designed, which abstracts multi-modal scene data into nodes and edges of graph and supports efficient organization of time, location and associations of multi-modal scene data. Then, a sparse matrix based method of in-memory representation and storage for spatiotemporal graph index is presented. Finally, taking the multi-dimensional tree index as an example, the index construction and multi-model query experiments are carried out. The experimental results show that the method is superior to the contrast method in several aspects, such as generation efficiency, query performance, and then supports real-time high-performance processing of dynamic and associated multi-modal scene data with low latency access.

Key words: spatiotemporal index, in-memory graph model, sparse matrix, dynamic and associated data, scene data organization