Indian Journal of Science and Technology

Abstract

Objective: To demonstrate the video event extraction system, combining the multicue optimized object detection with video semantic modelling. Methods/ Statistical Analysis: The literature established video semantic model has been recreated with a limited model. The multicue optimized object detector has been employed substituting the primary object detector in the system. The multicue object detector provides the inter object spatio-temporal cue information for event detection and aids to the semantic model. This similitude system establishes the proof of concept for the idea of incubating the multicue object identifier within a video extraction system for performance augmentation. The system had been tested for two different event extractions from a combination of six videos. Findings: The testing results had been presented for the two target events demonstrating the multicue object detector performing in frame by frame manner, for the test videos. The system performance measured and presented in terms of precision and recall indicates encouraging enhancements in the baseline system accuracy. Application/Improvement: The novelty of the work presented lies in clubbing the most advanced multicue optimized object detection technology with a semantic model based video extraction system, and thus bringing together the benefits of spatio-temporal cue method with a semantic model system.

Keywords: Event Detection, Event Extraction, Multicue Object Identifier, Spatio-Temporal Cue, Video Content Modelling