Ontology-based detection and identification of complex event of illegal parking using SPARQL and description logic queries

Abstract

Dynamic monitoring and management of traffic is a significant challenge which requires a mobile surveillance system which is becoming feasible with the evolution of 5G and related technologies. With the advent of 5G, huge amount of data can be transferred over long distance. Furthermore, the evolution of Multi-access Edge Computing (MEC) offers the processing near the edge, resulting in quick response which can cater to mobile surveillance. The mobile surveillance can be performed through a vehicle (Unmanned Aerial Vehicle (UAV)/mobile vehicle) which collects the multi-modal data and sends to nearest edge server over the 5G network. The server sends response quickly to the control center from where the appropriate action can be taken to mitigate its effect on the individuals in real-time. Furthermore, the analysis of multi-modal data captured by surveillance systems must be integrated and efficiently represented. Therefore, in this work, an ontology-based approach is proposed by integrating unstructured video data and structured sensor data for detecting and identifying the complex event of illegal parking. Identifying illegally parked vehicles needs dynamic and strict monitoring; which require the deployment of surveillance systems across the entire city. However, installing a video camera at each place is not feasible. Therefore, the surveillance thorough mobile vehicle is presented to identify illegal parked vehicle for traffic management over the 5G network. In addition, an ontology is developed to represent and reason over the roadside video data while integrating it with sensor data to form a knowledge graph. The knowledge graph retrieves semantic information about the events using SPARQL queries and Description Logics (DL) queries. Thus, a complete video is not required and the extracted information is stored in Resource Description Framework (RDF) format to reduce space requirement significantly. The approach is evaluated by recognizing the complex events of detecting illegally parked vehicles on a real-time dataset recorded using the experimental setup. Furthermore, the framework represents the surveillance video data of parked vehicles captured using the camera installed in a mobile vehicle to cover the road, including the vehicle information location, position, time, etc., as a knowledge graph that can be used later for further analysis.