A Light Vehicle License-Plate-Recognition System Based on Hybrid Edge-Cloud Computing.

Area of Science:

• Computer Science
• Artificial Intelligence
• Sustainable Development

Background:

• New-energy vehicles are rapidly growing globally, increasing demand for efficient systems.
• Current deep-learning-based license plate recognition (LPR) systems suffer from high latency and energy consumption.
• Over-reliance on cloud computing for LPR presents challenges in resource allocation and implementation.

Purpose of the Study:

• To propose an innovative Edge-LPR system for timely, effective, and energy-saving license plate recognition.
• To mitigate the limitations of cloud-based LPR systems by leveraging edge computing.
• To optimize LPR models for reduced resource consumption on edge devices.

Main Methods:

• Developed an Edge-LPR system utilizing edge computing and lightweight network models.
• Employed channel pruning to reconstruct the backbone layer and reduce network model parameters.
• Deployed network models on edge gateways using Intel's second-generation computing stick for direct license plate detection.

Main Results:

• The Edge-LPR system significantly reduces reliance on cloud computing resources.
• The optimized network model achieved a total parameter count of only 0.606 MB.
• Experimental validation on the CCPD dataset demonstrated a high accuracy rate of 97%.

Conclusions:

• The Edge-LPR system offers a reliable and effective solution for license plate recognition in new-energy vehicle applications.
• Edge computing integration with lightweight models provides an energy-efficient and low-latency alternative to traditional LPR systems.
• The proposed system demonstrates practical viability for real-time monitoring in environments like charging stations.