计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (13): 205-210.DOI: 10.3778/j.issn.1002-8331.2203-0276

• 图形图像处理 • 上一篇    下一篇

面向交通标识的二值语义嵌入学习方法

王少华,刘法胜,时柏营,刘兴波,聂秀山   

  1. 1.山东建筑大学 计算机科学与技术学院,济南 250101
    2.山东科技大学 电气与自动化工程学院,山东 青岛 266590
    3.山东建筑大学 交通工程学院,济南 250101
  • 出版日期:2023-07-01 发布日期:2023-07-01

Learning Binary Semantic Embedding for Traffic Signs

WANG Shaohua, LIU Fasheng, SHI Baiying, LIU Xingbo, NIE Xiushan   

  1. 1.School of Computer Science and Technology, Shandong Jianzhu University, Jinan 250101, China
    2.School of Electrical Engineering and Automation, Shandong University of Science and Technology, Qingdao, Shandong 266590, China
    3.School of Transportantion, Shandong Jianzhu University, Jinan 250101, China
  • Online:2023-07-01 Published:2023-07-01

摘要: 交通标识是交通基础设施的重要组成。智能交通系统中的自动驾驶车辆需要识别和理解交通标识,以确保其驾驶行为安全和遵守交通法规。目前,交通标识的识别大都采用深度神经网络方法,利用大量的训练样本对神经网络参数进行训练,获得对任务有利的特征表示,然而,海量的训练数据将带来较高的检索成本。针对以上问题,提出一种基于二值语义嵌入的大规模交通标识检索与识别方法BETS,该方法将标签信息和成对相似性信息嵌入到二值语义空间中,同时使用深度神经网络来进行哈希学习。实验结果表明,该方法可以有效提升大规模交通标识检索与识别的精度和准确性。

关键词: 交通标识图像, 二值码, 深度神经网络, 标签信息

Abstract: Traffic sign is an important component of traffic infrastructure. Autonomous vehicles in intelligent transportation systems need to recognize and understand traffic signs to ensure that their driving behavior is safe and comply with traffic laws and regulations. At present, deep neural networks are commonly used to recognize traffic signs. Deep neural network uses a large number of training samples to train neural network parameters and obtain the feature representation favorable to the task. However, massive training data will bring high retrieval cost. In view of the above problem, this paper presents a large-scale traffic signs based on binary semantic embedding retrieval and identification method named BETS. The proposed method embeds the label information and pairwise similarity into binary semantic space, and uses deep neural network in the Hash learning process. Experimental results show that the proposed method BETS can effectively improve the accuracy of large-scale traffic sign retrieval and recognition.

Key words: traffic sign image, binary code, deep neural network, label information