With the widespread deployment of ground, air and space sensor sources (internet of things or IoT, social networks, sensor networks), the integrated applications of real-time geospatial data from ubiquitous sensors, especially in public security and smart city domains, are becoming challenging issues. The traditional geographic information system (GIS) mostly manages time-discretized geospatial data by means of the Structured Query Language (SQL) database management system (DBMS) and emphasizes query and retrieval of massive historical geospatial data on disk. This limits its capability for on-the-fly access of real-time geospatial data for online analysis in real time. This paper proposes a hybrid database organization and management approach with SQL relational databases (RDB) and not only SQL (NoSQL) databases (including the main memory database, MMDB, and distributed files system, DFS). This hybrid approach makes full use of the advantages of NoSQL and SQL DBMS for the real-time access of input data and structured on-the-fly analysis results which can meet the requirements of increased spatio-temporal big data linking analysis. The MMDB facilitates real-time access of the latest input data such as the sensor web and IoT, and supports the real-time query for online geospatial analysis. The RDB stores change information such as multi-modal features and abnormal events extracted from real-time input data. The DFS on disk manages the massive geospatial data, and the extensible storage architecture and distributed scheduling of a NoSQL database satisfy the performance requirements of incremental storage and multi-user concurrent access. A case study of geographic video (GeoVideo) surveillance of public security is presented to prove the feasibility of this hybrid organization and management approach.

Public security incidents have been increasingly challenging to address with their new features, including large-scale mobility, multi-stage dynamic evolution, spatio-temporal concurrency and uncertainty in the complex urban environment, which require spatio-temporal association analysis among multiple regional video data for global cognition. However, the existing video data organizational methods that view video as a property of the spatial object or position in space dissever the spatio-temporal relationship of scattered video shots captured from multiple video channels, limit the query functions on interactive retrieval between a camera and its video clips and hinder the comprehensive management of event-related scattered video shots. GeoVideo, which maps video frames onto a geographic space, is a new approach to represent the geographic world, promote security monitoring in a spatial perspective and provide a highly feasible solution to this problem. This paper analyzes the large-scale personnel mobility in public safety events and proposes a multi-level, event-related organization method with massive GeoVideo data by spatio-temporal trajectory. This paper designs a unified object identify(ID) structure to implicitly store the spatio-temporal relationship of scattered video clips and support the distributed storage management of massive cases. Finally, the validity and feasibility of this method are demonstrated through suspect tracking experiments.

Vehicle re-identification plays an important role in cross-camera tracking and vehicle search in surveillance videos. Large variance in the appearance of the same vehicle captured by different cameras and high similarity of different vehicles with the same model poses challenges for vehicle re-identification. Most existing methods use a center proxy to represent a vehicle identity; however, the intra-class variance leads to great difficulty in fitting images of the same identity to one center feature and the images with high similarity belonging to different identities cannot be separated effectively. In this paper, we propose a sampling strategy considering different viewpoints and a multi-proxy constraint loss function which represents a class with multiple proxies to perform different constraints on images of the same vehicle from different viewpoints. Our proposed sampling strategy contributes to better mine samples corresponding to different proxies in a mini-batch using the camera information. The multi-proxy constraint loss function pulls the image towards the furthest proxy of the same class and pushes the image from the nearest proxy of different class further away, resulting in a larger margin between decision boundaries. Extensive experiments on two large-scale vehicle datasets (VeRi and VehicleID) demonstrate that our learned global features using a single-branch network outperforms previous works with more complicated network and those that further re-rank with spatio-temporal information. In addition, our method is easy to plug into other classification methods to improve the performance.