Record Details

CapsNet-based Precise and Rapid Traffic Sign Detection through AI in Adverse Environmental Scenarios

NOPR - NISCAIR Online Periodicals Repository

View Archive Info
 
 
Field Value
 
Title CapsNet-based Precise and Rapid Traffic Sign Detection through AI in Adverse Environmental Scenarios
 
Creator Kaur, Ravinder
Singh, Jitendra
Sharma, Swati
 
Subject Autonomous driving vehicles
Artificial intelligence
Intelligent transportation
Neural networks
Traffic sign detection
 
Description 989-1000
This research presents an innovative system for real-time recognition of traffic signs, leveraging Artificial Intelligence
(AI) to achieve high-performance identification, particularly under challenging conditions prevalent in traffic scenarios such
as occlusions, atmospheric haze, and noise. The proposed system employs an advanced Capsule Network (CapsNet)
algorithm for object recognition, trained extensively on a diverse dataset encompassing images of various traffic signs.
Notably, the system demonstrates concurrent detection capabilities for multiple traffic signals, accurately categorizing them
based on their respective classes. To address distortions caused by alterations in the camera's perspective, including rotation,
torsion, and elongation, the system effectively employs techniques ensuring precise alignment with the pertinent traffic sign.
Furthermore, pre-processing techniques are utilized to resolve ambiguity and distortions in complex traffic scenarios.
Empirical validation of the proposed methodology is conducted through experimentation with authentic traffic sign images
obtained from diverse environmental contexts. Comparative assessments across diverse datasets representing prominent
traffic sign domains affirm the efficacy of the proposed approach. The outcomes showcase a noteworthy precision level,
achieving a recognition accuracy of 99.16% for traffic signs. In contrast, conventional rule-based systems under identical
conditions exhibit accuracy rates ranging between 80–90%. The AI-driven system demonstrates real-time operational
feasibility, positioning it as a fitting candidate for applications in traffic management and intelligent transportation systems.
 
Date 2024-09-18T09:18:28Z
2024-09-18T09:18:28Z
2024-09
 
Type Article
 
Identifier 0022-4456 (Print); 0975-1084 (Online)
http://nopr.niscpr.res.in/handle/123456789/64555
https://doi.org/10.56042/jsir.v83i9.7388
 
Language en
 
Publisher NIScPR-CSIR,India
 
Source JSIR Vol.83(09) [September 2024]