基于计算机视觉与深度学习的车辆荷载识别研究进展  

Research Progress on Vehicle Load Identification Based on Computer Vision and Deep Learning

作　　者：李道恒 柳美玉 杨璐 LI Daoheng;LIU Meiyu;YANG Lu(Key Laboratory of Urban Security and Disaster Engineering of Ministry of Education,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学,城市与工程安全减灾教育部重点实验室,北京100124

出　　处：《综合运输》2025年第3期111-118,共8页China Transportation Review

基　　金：国家自然科学基金(52078011);北京市自然科学基金(8242004)。

摘　　要：近年来,交通运输行业迅速发展,超载车辆不断增加,严重影响公路桥梁的安全耐久水平,准确高效地识别车辆荷载信息对交通基础设施的运营、评估和维护具有重要意义。随着计算机视觉和深度学习技术的广泛应用,车辆荷载识别也得到进一步发展。本文从车辆荷载的直接识别及信息融合识别两个方面对目前的研究现状进行了分析,发现直接识别方法虽然能够取得一定的效果,但需要置于研究所处的具体环境下,而融合识别方法是目前的发展趋势;其次指出了当前车辆荷载识别领域的任务与难点,主要在于如何在准确可靠识别车辆荷载信息的前提下,寻求识别精度与速度之间的平衡;最后对相关研究工作进行总结,并讨论了未来的优化方向与发展趋势,本文认为将来应进一步优化视觉识别模型与算法,并与新兴技术不断交叉融合以提高现实复杂环境中的车辆荷载识别效果,使其更好地应用于工程实际。In recent years,the transportation industry has experienced rapid development,leading to a continuous increase in overloaded vehicles.This trend has significantly impacted the safety and durability levels of highway bridges.Accurate and efficient identification of vehicle load information is of paramount importance for the operation,assessment,and maintenance of transportation infrastructure.With the widespread application of computer vision and deep learning technologies,the field of vehicle load identification has seen further advancements.This paper analyzes the current research status in two aspects:direct identification of vehicle loads and information fusion identification.While direct identification methods have shown certain effectiveness,they are often context-dependent and influenced by specific environments.On the other hand,fusion identification methods emerge as the current developmental trend.The paper also identifies the tasks and challenges in the current field of vehicle load identification,emphasizing the need to balance recognition accuracy and speed while ensuring the accurate and reliable identification of vehicle load information.In conclusion,the paper summarizes the relevant research efforts and discusses future optimization directions and trends.It suggests that future advancements should focus on further optimizing visual recognition models and algorithms.Additionally,the paper advocates for continuous cross-disciplinary integration with emerging technologies to enhance the effectiveness of vehicle load identification in complex real-world environments,enabling better applications in engineering practices.

关 键 词：车辆荷载 计算机视觉 深度学习 卷积神经网络 

分 类 号：U446[建筑科学—桥梁与隧道工程]