Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems

The swift surge of video content depicts enormous challenges for intelligent information systems in extracting and searching video data. This article explores improvements achieved by introducing a Faster-RCNN hybrid model into video data extraction and search processes. We propose a novel methodolo...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhang Pei, Chen Zhengyi
Format: Article
Language:English
Published: De Gruyter 2025-08-01
Series:Nonlinear Engineering
Subjects:
Online Access:https://doi.org/10.1515/nleng-2025-0133
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849225137154424832
author Zhang Pei
Chen Zhengyi
author_facet Zhang Pei
Chen Zhengyi
author_sort Zhang Pei
collection DOAJ
description The swift surge of video content depicts enormous challenges for intelligent information systems in extracting and searching video data. This article explores improvements achieved by introducing a Faster-RCNN hybrid model into video data extraction and search processes. We propose a novel methodology combining Faster-RCNN with adaptive feature fusion and temporal coherence modeling to enhance object detection and tracking within video streams substantially. Using a large multi-megavideo dataset called VideoDiv-10K, which is made up of 10,000 videos from categories and others made available through public datasets like ImageNet VID and MOT16, this proposed model exhibited magnificent improvements. This model, when hybrid used with 500 h of content diversity, produced an average mean average precision of 0.891 and had an improved amount by 18.7% more than traditional Faster-RCNN. The model reduced its time for computation by 32.4% and had a massive leap in the accuracy in the search by 41.2%. Key to this performance were the temporal coherence module, which does a good job in capturing dependencies within sequential frames, and the adaptive feature fusion mechanism that dynamically integrates spatial and temporal features. Those were used to attain average F1-score in object detection of 0.937 and 0.912 object tracking on some video resolutions and frame rates. The results of the experiment clearly prove the ability of our hybrid model to process large video data, with significant gains in both accuracy and speed. We then provide better performance within applications such as surveillance, autonomous driving, and content-based video retrieval. Our model maintains a high level of scalability and does not degrade for huge datasets.
format Article
id doaj-art-dae10e4d821648b7bf42ee9c8653527a
institution Kabale University
issn 2192-8029
language English
publishDate 2025-08-01
publisher De Gruyter
record_format Article
series Nonlinear Engineering
spelling doaj-art-dae10e4d821648b7bf42ee9c8653527a2025-08-25T06:11:00ZengDe GruyterNonlinear Engineering2192-80292025-08-0114111210.1515/nleng-2025-0133Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systemsZhang Pei0Chen Zhengyi1Dance Academy, Sangmyung University, Seoul, 03015, KoreaDance Academy, Sangmyung University, Seoul, 03015, KoreaThe swift surge of video content depicts enormous challenges for intelligent information systems in extracting and searching video data. This article explores improvements achieved by introducing a Faster-RCNN hybrid model into video data extraction and search processes. We propose a novel methodology combining Faster-RCNN with adaptive feature fusion and temporal coherence modeling to enhance object detection and tracking within video streams substantially. Using a large multi-megavideo dataset called VideoDiv-10K, which is made up of 10,000 videos from categories and others made available through public datasets like ImageNet VID and MOT16, this proposed model exhibited magnificent improvements. This model, when hybrid used with 500 h of content diversity, produced an average mean average precision of 0.891 and had an improved amount by 18.7% more than traditional Faster-RCNN. The model reduced its time for computation by 32.4% and had a massive leap in the accuracy in the search by 41.2%. Key to this performance were the temporal coherence module, which does a good job in capturing dependencies within sequential frames, and the adaptive feature fusion mechanism that dynamically integrates spatial and temporal features. Those were used to attain average F1-score in object detection of 0.937 and 0.912 object tracking on some video resolutions and frame rates. The results of the experiment clearly prove the ability of our hybrid model to process large video data, with significant gains in both accuracy and speed. We then provide better performance within applications such as surveillance, autonomous driving, and content-based video retrieval. Our model maintains a high level of scalability and does not degrade for huge datasets.https://doi.org/10.1515/nleng-2025-0133faster-rcnnvideo data extractionobject detectiontemporal coherenceintelligent information systems
spellingShingle Zhang Pei
Chen Zhengyi
Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
Nonlinear Engineering
faster-rcnn
video data extraction
object detection
temporal coherence
intelligent information systems
title Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
title_full Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
title_fullStr Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
title_full_unstemmed Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
title_short Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
title_sort optimization effect of video data extraction and search based on faster rcnn hybrid model on intelligent information systems
topic faster-rcnn
video data extraction
object detection
temporal coherence
intelligent information systems
url https://doi.org/10.1515/nleng-2025-0133
work_keys_str_mv AT zhangpei optimizationeffectofvideodataextractionandsearchbasedonfasterrcnnhybridmodelonintelligentinformationsystems
AT chenzhengyi optimizationeffectofvideodataextractionandsearchbasedonfasterrcnnhybridmodelonintelligentinformationsystems