Scene Graph Driven Context Query Generation: A Focus on Diversity and Situation-Specific Queries
摘要
The rapid growth of IoT has produced vast data, but isolated systems limit comprehensive analysis and interoperability, hindering intelligent applications. To address this challenge, Context Management Platforms (CMPs) have emerged. However, current CMP performance evaluations often emphasize data ingestion, overlooking data digestion. This study introduces a novel scenario-based context query generation approach to evaluate data digestion performance of CMPs. Real-world scenes are modeled as context-aware scene graphs using ontology-based knowledge. This approach integrates ontological knowledge, sensor data, and real-world images to comprehensively represent urban road situations. We infer dynamic situations from the scene graph, forming the basis for generating diverse queries. A template-based query generation method ensures a range of queries with varying complexity. We evaluated the proposed approach using real-world datasets, demonstrating the query generation effectiveness and practicality. The findings show the correlation between the volume of queries and the scene complexity. We also analyzed computation time for query generation against scene complexity and assessed query relevance through weighted attribute coverage, demonstrating that our method effectively tailors queries to each scene’s dynamics. The proposed query generation approach lays a strong foundation for automating the evaluation of CMPs’ data retrieval/digestion capabilities.