A semantic segmentation-based algorithm and device for monitoring anomalies along the waterfront
By employing a semantic segmentation-based algorithm for monitoring anomalies along waterfronts, and utilizing a sampling module and an improved PVT model, combined with SoPhie and SlowFast models, the problem of low efficiency in waterway video surveillance has been solved, achieving efficient and accurate detection and early warning of abnormal targets.
CN116665041BActive Publication Date: 2026-07-03STATE GRID ZHEJIANG ELECTRIC POWER CO LTD +1
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- STATE GRID ZHEJIANG ELECTRIC POWER CO LTD
- Filing Date
- 2023-04-27
- Publication Date
- 2026-07-03
AI Technical Summary
Technical Problem
In existing technologies, frame-by-frame monitoring of waterways far from cities is inefficient, resulting in large amounts of redundant data and serious waste of resources.
Method used
A semantic segmentation-based algorithm for monitoring abnormal situations on the waterfront is adopted. The sampling module acquires sampling frames, uses the residual map of adjacent frames to determine scene changes, combines an improved PVT model for semantic segmentation, and uses SoPhie and SlowFast models for trajectory prediction and action recognition to output the warning level.
Benefits of technology
It improves the efficiency of waterway monitoring video data processing, reduces redundant data, enhances monitoring accuracy and early warning accuracy, and avoids safety accidents.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116665041B_ABST
Abstract
This invention discloses a waterfront anomaly monitoring algorithm and device based on semantic segmentation, comprising the following steps: S1: The sampling module performs sampling to obtain sampling frames; S2: The scene content is determined to have changed based on the residual map of two adjacent frames; S3: Semantic segmentation is performed on the current frame that has changed; S4: The abnormal target is determined to have appeared based on the target features obtained from semantic segmentation; S5: After an abnormal target appears, the warning level is determined using a trajectory prediction model and a recognition model. This invention reduces the workload of the abnormal target detection module and the abnormal behavior monitoring module, and improves the overall monitoring efficiency.
Need to check novelty before this filing date? Find Prior Art