融合多传感器与迁移学习的车牌识别方法研究  

作　　者：严格 

机构地区：[1]浙江同济科技职业学院,杭州311200

出　　处：《科技创新与应用》2023年第30期172-176,共5页Technology Innovation and Application

基　　金：省属高校一般科研项目(FRF22YB026)。

摘　　要：车牌精准定位和识别技术在现代交通管理、智能驾驶以及安防领域具有广泛的应用前景。然而,在复杂背景下,车牌的精准定位和识别效果并不是很理想。基于此,该文提出一种融合多传感器的车牌定位识别技术,首先采用激光雷达技术获取车牌三维点云数据,并与图像传感器采集的车牌图像进行双重验证,实现车牌精准定位。同时根据图像传感器获取的车牌信息进行车牌字符分割,并建立字符数据集。最后利用迁移学习思想,采用VGGNet网络搭建车牌字符图像分类模型进行车牌定位识别。实验结果表明,该文提出方法的车牌定位平均准确率为90%,车牌识别平均准确率为92%。该方法在复杂背景下具有较好的稳定性和准确性,为车牌定位与识别领域提供一种新的解决途径。Accurate vehicle license plate localization and recognition technology hold extensive potential for applications in modern traffic management,intelligent driving,and security domains.However,achieving precise license plate localization and recognition results remains challenging,especially in complex backgrounds.To address this issue,this paper proposes a multi-sensor integrated approach for license plate localization and recognition.Initially,laser radar technology is employed to capture three-dimensional point cloud data of license plates,which is then cross-verified with license plate images obtained from image sensors to achieve accurate plate localization.Simultaneously,license plate character segmentation is conducted based on the information acquired from image sensors,and a character dataset is established.Lastly,leveraging the concept of transfer learning,a VGGNet network is constructed for license plate character image classification,facilitating license plate localization and recognition.Experimental outcomes demonstrate an average localization accuracy of approximately 90%and an average recognition accuracy of 92%for the proposed method.This technique exhibits robust stability and precision in complex backgrounds,offering a novel approach for advancements in the field of license plate localization and recognition.

关 键 词：激光雷达 车牌定位 字符分割 迁移学习 车牌识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]