基于深度学习的单阶段车辆检测算法综述

Review of one-stage vehicle detection algorithms based on deep learning

随着基于深度学习的车辆检测技术的更新迭代,它在智能交通中发挥着越来越重要的作用。单阶段目标检测网络因其检测速度快的优点,被广泛地应用于监控视频中的车辆实时检测。为了综合分析和了解现今各种深度学习网络在车辆检测中应用情况,先介绍了当前深度学习中各类目标检测方法在车辆检测中的应用,之后简介了当前常用的单阶段目标检测算法,详细阐述了实时车辆检测中各类单阶段检测算法的实际应用状况,列举了这些算法的优点和不足。最后,简单介绍了车辆检测相关数据集和评价标准,对目前车辆检测中待解决的问题、未来待改进的方向进行了分析和讨论,为车辆检测的发展方向提供了思路。

With the update and iteration of vehicle detection technology based on deep learning,it plays an increasingly important role in intelligent transportation.Single-stage target detection network was widely used in real-time detection of vehicles in surveillance video due to its fast detection speed.To comprehensively analyze and understand the applications of various deep learning networks in vehicle detection today,the application of various target detection methods in current deep learning was introduced into vehicle detection,and then the current commonly used single-stage target detection algorithm was presented,and actual application status of various single-stage detection algorithms were elaborated in real-time vehicle detection.The advantages and disadvantages of these algorithms were listed.Finally,the data sets and evaluation criteria of vehicle detection were briefly introduced,and the problems to be solved in vehicle detection and the direction to be improved in the future were analyzed and discussed,which provides the development direction of vehicle detection.