The Deep Learning-Based Semantic Cross-Modal Moving-Object Moment Retrieval System

This paper presents a high-efficiency deep learning-based semantic cross-modal moving-object moment retrieval system that is applied to surveillance video system for greatly reducing both human resource cost and time-consuming. The originality of our work is to improve the fusion processing of different modal features to reduce the semantic gap problem and design a moment location module to avoid various interferences and sampling restrictions to meet the specific conditions for achieving more efficient retrieval than other methods.