A reinforcement learning-based intelligent detection triggering method, medium, and device for H.264 video.

By parsing H.264 video streams to extract encoding domain metadata and constructing multi-dimensional temporal state vectors, and using reinforcement learning to optimize the decision model, the problem of computational resource consumption caused by traditional decoding is solved, achieving low-cost and highly robust video detection triggering, which is suitable for edge computing and server-side scenarios.

CN122093583APending Publication Date: 2026-05-26NINGBO BEILUN FIRST CONTAINER TERMINAL CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
NINGBO BEILUN FIRST CONTAINER TERMINAL CO LTD
Filing Date
2026-04-10
Publication Date
2026-05-26

Smart Images

  • Figure CN122093583A_ABST
    Figure CN122093583A_ABST
Patent Text Reader

Abstract

This application discloses a method, medium, and device for intelligent detection triggering of H.264 video based on reinforcement learning. The method includes: real-time parsing of the H.264 video bitstream to directly extract metadata from the coding domain; constructing a multi-dimensional temporal state vector that integrates motion, coding, and temporal features based on this metadata; and inputting this vector into a pre-trained decision model to output a trigger command. The decision model is based on the Q-Learning algorithm, which selects the action to trigger or skip detection based on the current state by establishing and iteratively updating a Q-value table, and calculates a reward based on the action execution result to continuously optimize the decision strategy. This invention utilizes metadata from the coding domain for intelligent decision-making, fundamentally solving the computational bottleneck problem caused by full-frame decoding in multi-stream video AI analysis, and significantly reducing system computational resource consumption while ensuring detection accuracy.
Need to check novelty before this filing date? Find Prior Art