基于机器视觉的车辆检测与参数识别研究进展

Research Advances on Vehicle Parameter Identification Based on Machine Vision

近年来公路交通运输快速增长,交通车辆的快速准确检测与识别对智能交通系统和交通基础设施运维具有重要意义。随着机器视觉和深度学习技术的迅速发展及其在目标检测领域的广泛应用,车辆目标检测和参数识别也取得新的突破。该文从车辆参数的识别方法和应用研究两方面梳理了机器视觉和深度学习在车辆检测与参数识别领域的研究现状、最新研究成果和未来发展趋势。在识别方法方面,将车辆检测方法分为3类:运动目标检测方法、目标实例检测方法和细粒度检测方法,系统总结了这3类方法的基本原理和各自特点。在应用研究方面,详细综述了基于机器视觉的车辆检测方法在车辆参数识别中的应用现状,主要包括车辆类别、车辆时空参数、车辆重量参数识别以及车辆多参数识别系统。最后对基于机器视觉和深度学习的车辆参数识别研究进行了归纳总结,并讨论了当前存在的挑战和未来可能的发展趋势。研究表明,对于不同的环境条件和车辆参数,应根据实际需要和各算法特点选择合适的车辆检测方法。目前方法仍局限于单参数或少量参数的独立检测,且识别精度和效率难以同时满足。后续研究应注重与新技术的融合,提高在现实复杂环境下车辆参数识别的精度、效率、鲁棒性及全面性,以使其更好地应用于工程实际。

With the rapid growth of highway transportation in recent years,the accurate identification of vehicle parameters is of great significance to the development of intelligent transportation and the operation and maintenance of transportation infrastructure.The rapid development and popularity of machine vision and deep learning in moving object detection have achieved new advances in vehicle parameter identification.The present study aimed to review the current status,new advances,and future trends in vehicle parameter identification based on machine vision and deep learning technologies from the two aspects of detection algorithm and applied research.Regarding detection methods,this paper introduced the basic principles,pros,and cons of these methods by classifying them into three categories:moving object detection,instance object detection,and fine-grained object detection.Research on the identification of vehicle parameters based on machine vision and deep learning was reviewed in detail,including the vehicle type,vehicle spatio-temporal parameters,vehicle weight parameters,and multiple parameter identification systems.Finally,this paper summarized research on vehicle parameters based on machine vision and deep learning,discussing the current challenges and possible future trends.The research demonstrates that suitable vehicle detection methods should be selected according to the actual requirements and the characteristics of each algorithm for different environmental conditions and vehicle parameters.At present,the detection method is still limited to single-vehicle parameters or the independent detection of several parameters.As identification accuracy and efficiency are difficult to meet simultaneously,future research should focus on the integration of new technologies to improve the accuracy, efficiency, robustness, and comprehensive vehicle parameter identification in complex environments for better application in practice.