基于多信息融合的高速收费站拥堵检测算法  

Congestion Detection Algorithm of Highway Toll Station Based on Multi-Information Fusion

作　　者：王晓龙 江波 罗润书 安国成 WANG Xiaolong;JIANG Bo;LUO Runshu;AN Guocheng(Artificial Intelligence Research Institute,Shanghai Huaxun Network System Co.,Ltd.,Chengdu 610074,Sichuan,China;School of Computer Science and Technology,Hangzhou Dianzi University,Hangzhou 310018,Zhejiang,China;The 32nd Research Institute of China Electronics Technology Group Corporation,Shanghai 201808,China)

机构地区：[1]上海华讯网络系统有限公司行业数智事业部,四川成都610074 [2]杭州电子科技大学计算机学院,浙江杭州310018 [3]中国电子科技集团公司第三十二研究所,上海201808

出　　处：《计算机工程》2025年第5期377-386,共10页Computer Engineering

基　　金：国家重点研发计划(2023YFC3006700)。

摘　　要：在高速收费站拥堵检测场景中,遮挡、阴影、景深等因素严重影响检测算法的精度,基于单信息的拥堵检测算法无法准确反映收费站真实拥堵情况,对此,提出一种高速收费站多信息融合拥堵检测算法。首先采用目标检测算法检测目标区域车辆,利用图像空间特征计算车道占用率;其次利用Deep SORT目标跟踪算法与滑动平衡机制进行车流量综合估计,减少因为遮挡和阴影导致的车辆误检、漏检问题;最后利用基于光流信息熵的车速估计方法,降低因景深变化导致的拥堵状态检测误差。通过融合3种不同维度信息得到拥堵指数,拥堵指数聚类为5种拥堵类别以判断真实拥堵状态,从而实现高速收费站的拥堵检测。实验结果表明,在高速收费站拥堵数据集上,采用多信息融合的拥堵检测算法准确率达到90.4%,能够实现对收费站拥堵状态的准确检测。At highway toll stations,factors such as occlusion,shadows,and the depth of field significantly impact the accuracy of congestion detection.Moreover,relying solely on a single parameter for congestion detection cannot accurately reflect the actual congestion situation at toll stations.To address these challenges,this paper proposes a multistep approach.First,a target detection algorithm is employed to identify vehicles within the designated area.Spatial features extracted from the images of these vehicles are used to calculate the lane occupancy rate.Second,comprehensive vehicle flow is estimated utilizing the Deep SORT target tracking algorithm and sliding balance mechanism to mitigate false and missed detections caused by occlusion and shadowing effects.Furthermore,an optical flow-based vehicle speed estimation method is utilized to reduce congestion detection errors resulting from changes in the depth of field.Finally,through the fusion of three distinct dimensions of information(target presence,lane occupancy rate,and vehicle speed),a congestion index is obtained.The values of this index can be clustered into five categories to determine the real-time congestion state at highway toll stations.Experimental are conducted on a congested dataset collected from high-speed toll stations.The results demonstrate that the proposed multi-information fusion-based congestion detection algorithm achieves an accuracy rate of 90.4%,enabling the precise identification of traffic congestion at toll stations.

关 键 词：多信息融合 目标检测 目标跟踪 车速估计 拥堵检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]