Gaussian Kernels Based Network for Multiple License Plate Number Detection in Day-Night Images
Document Type
Conference Item
Publication Date
1-1-2023
Abstract
Detecting multiple license plate numbers is crucial for vehicle tracking and re-identification. The reliable detection of multiple license plate numbers requires addressing the challenges like image defocusing and varying environmental conditions like illumination, sunlight, shadows, weather conditions etc. This paper aims to develop a new approach for multiple license plate number detection of different vehicles in day and night scenarios. The proposed work segments the vehicle region containing license plate numbers based on a multi-column convolutional neural network and iterative clustering to reduce the background challenges and the presence of multiple vehicles. To address challenges of font contrast variations and text-like objects in the background, the proposed work introduces the Gaussian kernels that represent a text pixel distribution to integrate with a proposed deep learning model for detection, Experimental results on benchmark datasets of day and night license plate number show that the proposed model is effective and outperforms the existing methods.
Keywords
Text detection, Vehicle detection, Text segmentation, Deep learning, Gaussian kernels, Multiple license plate number detection
Divisions
fsktm
Funders
IDEAS-Technology Innovation Hub grant, Indian Statistical Institute, Kolkata, India
Volume
14191
Publisher
SPRINGER INTERNATIONAL PUBLISHING AG
Publisher Location
GEWERBESTRASSE 11, CHAM, CH-6330, SWITZERLAND
Event Title
Document Analysis and Recognition - ICDAR 2023, Pt V
Event Location
San Jose, California
Event Dates
21-26 August 2023
Event Type
conference
Additional Information
17th International Conference on Document Analysis and Recognition (ICDAR), San Jose, CA, AUG 21-26, 2023