Enhancement of license plate recognition performance using Xception with Mish activation function.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems (ITS)

Background:

• Intelligent Transportation Systems (ITS) rely on effective Vehicle Licence Plate Recognition (VLPR).
• Existing VLPR methods struggle with environmental complexities like motion blur, low resolution, and varied lighting.
• There is a need for robust VLPR solutions that perform well under diverse and challenging conditions.

Purpose of the Study:

• To develop an effective and robust solution for recognizing characters on license plates under complex environmental conditions.
• To address limitations in current VLPR methods, particularly concerning motion blur, low resolution, and adverse weather/lighting.
• To enhance the capabilities of Intelligent Transportation Systems through improved license plate recognition.

Main Methods:

• A novel approach combining Generative Adversarial Networks (GAN) with a Discrete Cosine Transform (DCT) Discriminator (DCTGAN) for joint super-resolution and deblurring.
• License Plate (LP) detection using the Improved Bernsen Algorithm (IBA) integrated with Connected Component Analysis (CCA).
• Character recognition on LPs utilizing an Xception model with transfer learning, notably without character segmentation.

Main Results:

• The proposed DCTGAN method effectively handles motion blur and various image complexities in license plates.
• The integrated system demonstrates robust performance across diverse datasets and challenging conditions, including low resolution, poor lighting, and adverse weather.
• The approach achieves superior results compared to current state-of-the-art methods in both accuracy and quality.

Conclusions:

• The developed VLPR system offers a significant advancement in recognizing license plates under difficult environmental factors.
• The combination of DCTGAN, IBA, CCA, and transfer learning-based Xception model provides a powerful and versatile solution for ITS applications.
• This research contributes to more reliable and efficient intelligent transportation systems by overcoming key challenges in license plate recognition.