Title: Research on multimedia video technology based on the DR-DVC algorithm
Authors: Jing Xiao
Addresses: College of Tourism and Financial Media, Xi'an Siyuan University, Xi'an City, Shaanxi Province, 710000, China
Abstract: In order to solve the problem of motion detection and event description in multimedia video, this paper proposes a dense event description method (dense video caption based on descriptiveness regression, DR-DVC) based on descriptive regression, and puts forward the corresponding evaluation indexes, including bleu-1, bleu-4, cider, metor and srough-l. DR-DVC optimises the original C3D network through principal component analysis, optimises timing action detection with descriptive regression and prior knowledge, and optimises the event description algorithm through an attention mechanism. The results show that with the increase of tiou, the recall rates of the four sequential action description generation algorithms show a significant downward trend. The area under the curve of the TAP-2 model is 57.61. After introducing the language generation model, the DR-DVC intensive time description generation algorithm will get better candidate proposals. The research results have extremely prominent value in multimedia video action detection and event description.
Keywords: DR-DVC; multiscale features; joint optimisation; video description; sequential action detection; principal component analysis.
DOI: 10.1504/IJCSM.2022.128684
International Journal of Computing Science and Mathematics, 2022 Vol.16 No.4, pp.327 - 339
Received: 13 Apr 2022
Accepted: 23 Sep 2022
Published online: 01 Feb 2023 *