Integration of Multi-Camera Video Moving Objects and GIS

Yujia Xie, Meizhen Wang, Xuejun Liu, Bo Mao, Feiyue WangISPRS Int. J. Geo Inf 2019 · DOI

TL;DR

This paper tackles the challenge of efficiently managing and analyzing vast amounts of surveillance video data by integrating multi-camera video moving objects (MCVO) with Geographic Information Systems (GIS). They propose a novel conceptual model and a prototype system (GIS-MCVO) that enhances video retrieval, analysis, and visualization by spatially contextualizing moving objects within a 3D geographic scene.

What Problem Does This Paper Address?

The proliferation of surveillance cameras worldwide generates an overwhelming volume of video data. Traditional methods of manually browsing and analyzing this data are inefficient and struggle to capture the spatial relationships between different camera feeds and the moving objects within them. This makes complex tasks like spatio-temporal behavior analysis and scene simulation difficult. Existing systems often focus on integrating video images with GIS, but this paper highlights the need to specifically integrate moving objects from multiple cameras into GIS to overcome limitations in manual browsing, retrieval, and analysis.

Methods

The core of this work lies in the integration of multi-camera video moving object (MCVO) extraction with GIS. The process involves several key steps:

  • MCVO Extraction: This involves three stages:
  • Object Detection: Identifying moving objects within individual video frames.
  • Object Tracking: Following the movement of detected objects across consecutive frames.
  • Cross-Camera Recognition: Identifying the same object appearing in multiple camera views to establish spatio-temporal associations.
  • Spatialization and Mapping: A homography matrix method is used to establish a mapping between image coordinates and geospatial coordinates. This allows for the georeferencing of video objects.
  • Fusion Model: The paper proposes a conceptual fusion model for MCVO and GIS, emphasizing the importance of foreground projection for expressing image-scene correlation.
  • GIS-MCVO Prototype System: A system architecture is designed and implemented, featuring a service-oriented approach with layers for function, data, service, business, and representation.
  • Key Functions: The system includes modules for:
  • Moving object extraction and data organization.
  • Video spatialization (camera calibration and mapping).
  • Virtual scene generation.
  • Spatial-temporal analysis of moving objects.
  • Fusion representation (visualizing MCVO within the GIS).

Datasets

Not discussed in the paper. The paper focuses on the methodology and system design, with examples likely generated from simulated or real-world surveillance footage without explicit dataset mentions.

Key Insights and Ideas

  • Moving Objects as the Core: The paper argues that focusing on moving objects, rather than just video frames, is crucial for efficient surveillance analysis.
  • Spatio-Temporal Pipeline (STP): This is introduced as a fundamental unit for describing MCVO information, enabling unified representation across multiple cameras.
  • Foreground Projection for Fusion: The authors found foreground projection to be the most suitable method for integrating MCVO and geographic scenes, balancing scene correlation with object prominence.
  • Video Compression at the Object Level: By storing only sub-graphs and trajectories of moving objects (not backgrounds), significant data compression is achieved, moving from an image-level to an object-level representation.
  • Trajectory Deduction in Blind Zones: The system can estimate and supplement object trajectories even in areas not covered by any camera, leveraging spatial relationships and geographic models.
  • GIS-Enhanced Retrieval: The GIS-MCVO system allows for both camera-by-camera and global trajectory-based retrieval of moving objects, overcoming limitations of traditional image-space retrieval.
  • Video Synopsis for Efficiency: The system generates video synopses that condense spatio-temporal information of multiple moving objects, significantly reducing viewing time.

Key Results

  • Improved Visualization: The GIS-MCVO system provides a more intuitive and effective way to visualize video information by integrating moving objects into a 3D geographic scene, overcoming the limitations of grid-based interfaces.
  • Efficient Data Compression: The object-level compression achieved compression rates in the order of 10⁻² to 10⁻⁴, significantly reducing storage requirements.
  • Enhanced Retrieval: The system demonstrates the ability to perform both local and global retrieval of moving objects based on their trajectories, leading to faster and more accurate results.
  • Reduced Viewing Time: Video synopsis generation effectively reduces the time needed to review surveillance footage by concentrating key events.
  • Qualitative Benefits: The integration of MCVO and GIS leads to improved efficiency in expressing video information, rapid browsing of objects across multiple cameras, and better understanding of spatio-temporal behaviors.

Limitations and Future Work

  • Loss of Visual Information: Acknowledged limitation is the abandonment of some visual image information of video objects during the compression process.
  • Uncertainty in Detection: The system can be affected by missed or false detections of video objects.
  • Visual Blind Zones: While trajectory deduction is implemented, visual blind zones between cameras can still decrease the accuracy of extracted MCVO trajectories.
  • Future Work: The authors suggest integrating additional data sources like GPS real-time positioning, remote sensing images, and lyric data to further enhance geographic video understanding and analysis. They also envision the prototype system serving as a platform for interactive monitoring and analysis with added functionalities.

Summary updated

BibTeX
@article{xie2019integration,
  author = {Yujia Xie and Meizhen Wang and Xuejun Liu and Bo Mao and Feiyue Wang},
  title = {{Integration of Multi-Camera Video Moving Objects and GIS}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2019},
  doi = {10.3390/ijgi8120561},
}
Figure