LF-RTMDet: an instance segmentation algorithm for real-time detection of water-filled barriers

Water-filled barriers are commonly used traffic control devices, typically employed in road maintenance and regulation scenarios. However, their configuration, formed by interconnected rods, creates a continuous barrier that poses an obstacle for autonomous vehicles. To address this, we selected the...

Full description

Saved in:
Bibliographic Details
Published in:The Visual computer Vol. 41; no. 10; pp. 7939 - 7950
Main Authors: Wang, JiaHao, Wang, Yongqiang, Zhou, Congling, Huang, Jiawei
Format: Journal Article
Language:English
Published: Heidelberg Springer Nature B.V 01.08.2025
Subjects:
Accuracy
Algorithms
Autonomous vehicles
Complexity
Control equipment
Deep learning
Inference
Instance segmentation
Kitchenware
Modules
Real time
Road maintenance
Roads & highways
Semantics
Sensors
Traffic control
ISSN:0178-2789, 1432-2315
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Water-filled barriers are commonly used traffic control devices, typically employed in road maintenance and regulation scenarios. However, their configuration, formed by interconnected rods, creates a continuous barrier that poses an obstacle for autonomous vehicles. To address this, we selected the RTMDet instance segmentation model as the baseline and proposed a lightweight, fine-grained RTMDet instance segmentation model, called LF-RTMDet. LF-RTMDet incorporates the GhostConv module for lightweight architecture, adds the CARAFE upsampling module to enhance fine-grained segmentation capabilities, and uses the PSA attention mechanism to improve WFB recognition in complex environments. Experimental results show a significant reduction in both the number of parameters and computational complexity, while inference accuracy is significantly improved. Finally, the LF-RTMDet model, after INT8 quantization, achieves a real-time inference speed of 32 FPS on the Jetson Orin Nano, with minimal accuracy loss, meeting the requirements for real-time detection.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0178-2789
1432-2315
DOI:10.1007/s00371-025-03847-3