面向跨相机鸟瞰视角道路对齐的地平线检测方法

doi:10.11772/j.issn.1001-9081.2025060733

《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (6): 1956-1964.DOI: 10.11772/j.issn.1001-9081.2025060733

• 多媒体计算与计算机仿真 • 上一篇下一篇

面向跨相机鸟瞰视角道路对齐的地平线检测方法

王伟, 刘佳欣(), 向婉妮, 崔华, 李阳光

长安大学信息工程学院，西安 710064

收稿日期:2025-07-04 修回日期:2025-09-22 接受日期:2025-09-29 发布日期:2025-10-13 出版日期:2026-06-10
通讯作者: 刘佳欣
作者简介:王伟（1984—），男，江苏徐州人，副教授，博士，CCF会员，主要研究方向：三维检测
向婉妮（1998—），女，陕西安康人，博士研究生，主要研究方向：计算机视觉
崔华（1977—），女，河南南阳人，教授，博士，主要研究方向：智慧交通
李阳光（2003—），男，湖北武汉人，主要研究方向：目标检测。
第一联系人：刘佳欣（2001—），女，河北廊坊人，硕士研究生，主要研究方向：三维检测
基金资助:
国家重点研发计划项目子课题(2024YFB2505701);国家自然科学基金面上项目(12171234);国家自然科学基金面上项目(62576050)

Horizon detection method for cross-camera bird’s-eye view road alignment

Wei WANG, Jiaxin LIU(), Wanni XIANG, Hua CUI, Yangguang LI

School of Information Engineering，Chang’an University，Xi’an Shaanxi 710064，China

Received:2025-07-04 Revised:2025-09-22 Accepted:2025-09-29 Online:2025-10-13 Published:2026-06-10
Contact: Jiaxin LIU
About author:WANG Wei， born in 1984， Ph. D.， associate professor. His research interests include 3D detection.
XIANG Wanni， born in 1998， Ph. D. candidate. Her research interests include computer vision.
CUI Hua， born in 1977， Ph. D.， professor. Her research interests include intelligent transportation.
LI Yangguang， born in 2003. His research interests include object detection.
First author contact:LIU Jiaxin， born in 2001， M. S. candidate. Her research interests include 3D detection.
Supported by:
National Key Research and Development Program of China （Sub-project）(2024YFB2505701);General Program of National Natural Science Foundation of China(12171234)

摘要/Abstract

摘要：

针对高速公路广泛部署的监控相机单相机视野有限，难以实现大范围连续感知的问题，提出跨相机鸟瞰视角（BEV）的道路几何对齐任务，以提升场景一致性与完整性。然而，由于多相机图像间存在视角差异与结构错位，该任务面临较大挑战。地平线作为全局几何先验能够统一视角差异。针对它的检测易受遮挡和环境干扰，制约对齐精度的问题，提出面向跨相机BEV道路对齐的路侧地平线检测方法RoadHoriNet （Road Horizon detection Network）。首先，通过透视变换与包围框裁剪进行数据增强；其次，引入钻石空间表示缓解消失点学习不稳定；再次，结合感受野注意力卷积（RFAConv）与动态上采样（DySample）提升特征表达与重建精度；最后，设计几何一致性损失函数，加强地平线检测的方向与位置约束。在公开数据集BrnoCompSpeed上的实验结果表明，RoadHoriNet 在像素误差（5.166%）和角度误差（0.032 5°）方面表现最佳，检测精度达到94.834%。与自适应地平线检测方法相比，像素误差减小了4.815个百分点，角度误差降低了0.019 4°；在跨相机BEV道路几何对齐任务中，经过RoadHoriNet方法修正后，相对对齐精度达99.129%以上，验证了该方法在交通环境中的实用性与推广潜力。可见，所提出的地平线检测方法能够为相机姿态归一化与坐标统一提供稳定的几何先验，有效提升跨相机BEV道路几何对齐的相对对齐精度与鲁棒性。

关键词: 地平线检测, 钻石空间, 几何一致性损失, 多相机对齐, 感受野注意力卷积, 动态上采样

Abstract:

To address the problem that the limited field of view of single camera of widely deployed highway surveillance cameras makes it difficult to achieve large-scale continuous perception， a cross-camera Bird’s-Eye View （BEV） road geometric alignment task was proposed to improve scene consistency and completeness. However， this task faces challenges due to the perspective differences and structural misalignments among multi-camera images. The horizon， as a global geometric prior， can unify these perspective differences， but its detection is easily affected by occlusion and environments， limiting alignment accuracy. To solve this problem， a horizon detection method for cross-camera BEV road alignment， named RoadHoriNet （Road Horizon detection Network）， was proposed. Firstly， perspective transformation and bounding box cropping were applied for data augmentation. Secondly， a diamond space representation was introduced to alleviate instability in vanishing-point learning. Thirdly， Receptive-Field Attention Convolution （RFAConv） and upsampling by Dynamic Sampling （DySample） were used to enhance feature representation and reconstruction accuracy. Finally， a geometric consistency loss function was designed to enhance the constraints of the orientation and position of horizon detection. Experimental results demonstrate that on the BrnoCompSpeed dataset， RoadHoriNet achieves a pixel error of 5.166%， an angle error of 0.032 5°， and a detection accuracy of 94.834%， while reducing the pixel error by 4.815 percentage points and the angle error by 0.019 4° compared with the adaptive horizon detection method. In the task of cross-camera BEV road geometry alignment， the relative alignment accuracy of RoadHoriNet reaches at least 99.129% after being corrected by the RoadHoriNet method， demonstrating its practicality and generalization potential in real-world traffic environments. It can be seen that RoadHoriNet provides a stable geometric prior for camera pose normalization and multi-camera coordinate unification， improving the relative alignment accuracy and robustness of cross-camera BEV road geometric alignment significantly.

Key words: horizon detection, diamond space, geometric consistency loss, multi-camera alignment, Receptive-Field Attention Convolution (RFAConv), Dynamic Upsampling (DySample)

中图分类号:

TP391.4

王伟, 刘佳欣, 向婉妮, 崔华, 李阳光. 面向跨相机鸟瞰视角道路对齐的地平线检测方法[J]. 计算机应用, 2026, 46(6): 1956-1964.

Wei WANG, Jiaxin LIU, Wanni XIANG, Hua CUI, Yangguang LI. Horizon detection method for cross-camera bird’s-eye view road alignment[J]. Journal of Computer Applications, 2026, 46(6): 1956-1964.

图/表 13

图1 本文方法的整体流程

Fig. 1 Overall flow of proposed method

图2 数据预处理

Fig. 2 Data preprocessing

图3 图像-钻石空间映射图

Fig. 3 Image-to-diamond space mapping diagram

图4 地平线检测流程及网络结构

Fig. 4 Horizon detection flow and network structure

图5 相机标定的和跨场景的相机模型

Fig. 5 Camera calibration and cross-scene camera models

图6 地平线几何约束下的跨场景图像对齐

Fig. 6 Cross-scene image alignment under horizon geometric constraints

图7 数据集图像示例

Fig. 7 Dataset image samples

表1 RoadHoriNet的消融实验结果

Tab. 1 Ablation experiment results of RoadHoriNet

实验名	感受野注意力卷积	动态上采样	地平线一致性损失	像素误差/%	角度误差/%	推理时延/ms
基准实验	×	×	×	11.532	0.069 8	2.504
实验1	√	×	×	7.317	0.045 0	1.917
实验2	×	√	×	6.421	0.053 1	1.118
实验3	×	×	√	5.951	0.009 9	1.504
实验4	√	√	×	5.539	0.013 5	1.320
实验5	√	×	√	6.705	0.036 5	1.831
实验6	×	√	√	5.930	0.025 7	1.244
实验7	√	√	√	5.166	0.032 5	1.336

表2 不同方法的实验误差对比

Tab. 2 Comparison of experimental errors of different methods

方法	像素误差/%	角度误差/（°）
GHLD^［6］	16.245	0.967 4
BHLCM^［28］	18.876	0.100 2
Kocur等^［14］方法	10.374	0.057 8
王伟等^［15］方法	9.981	0.051 9
RoadHoriNet	5.166	0.032 5

表3 RoadHoriNet在不同数据集上的泛化表现

Tab. 3 Generalization performance of RoadHoriNet on different datasets

数据集	像素误差/%	角度误差/（°）
BrnoCompSpeed	5.166	0.032 5
自制高速公路数据集	2.443	0.011 2
RCooper	6.660	0.057 3
自制城市道路数据集	10.216	0.056 0

图8 场景泛化验证结果

Fig. 8 Scene generalization verification results

图9 跨场景BEV道路几何对齐中的地平线约束对精度的影响

Fig. 9 Impact of horizon constraints on accuracy in cross-scene BEV road geometric alignment

表4 高速公路跨场景下的BEV道路几何对齐的相对对齐精度

Tab. 4 Relative alignment accuracy of cross-scene BEV road geometric alignment in highway scenarios

场景 1	场景2	真实距离/m	无地平线约束		有地平线约束
场景 1	场景2	真实距离/m	计算得到的距离/m	Acc_align/ %	计算得到的距离/m	Acc_align/ %
点1	点1'	60.000	59.422	99.037	59.932	99.887
点2	点2'	90.000	88.815	98.683	89.612	99.569
点3	点3'	120.000	116.647	97.206	119.300	99.417
点4	点4'	150.000	144.433	96.289	149.039	99.359
点5	点5'	120.000	116.248	96.873	118.955	99.129
点6	点6'	210.000	220.770	94.871	208.668	99.367

参考文献 27

[1]	黄德启，黄海峰，黄德意，等. BEV感知学习在自动驾驶中的应用综述［J］. 计算机工程与应用， 2025， 61（6）： 1-21.
	HUANG D Q， HUANG H F， HUANG D Y， et al. Review of application of BEV perceptual learning in autonomous driving［J］. Computer Engineering and Applications， 2025， 61（6）： 1-21.
[2]	HARTLEY R， ZISSERMAN A. Multiple view geometry in computer vision［M］. 2nd ed. Cambridge： Cambridge University Press， 2004.
[3]	ZHAI M， WORKMAN S， JACOBS N. Detecting vanishing points using global image context in a non-Manhattan world［C］// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2016： 5657-5665.
[4]	FU J， LI F， ZHAO J. Real-time infrared horizon detection in maritime and land environments based on hyper-Laplace filter and convolutional neural network［J］. IEEE Transactions on Instrumentation and Measurement， 2023， 72： No.5016513.
[5]	JEONG C， YANG H S， MOON K. A novel approach for detecting the horizon using a convolutional neural network and multi-scale edge detection［J］. Multidimensional Systems and Signal Processing， 2019， 30（3）： 1187-1204.
[6]	XU Y， YAN H， MA Y， et al. Graph-based horizon line detection for UAV navigation［J］. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing， 2021， 14： 11683-11698.
[7]	LI F， ZHANG J， SUN W， et al. Sea-sky line detection using gray variation differences in the time domain for unmanned surface vehicles［J］. Signal， Image and Video Processing， 2021， 15（1）： 139-146.
[8]	LI D， WU Z， ZUO Z， et al. Skyline detection method based on offset heatmap network［C］// Proceedings of the IEEE 4th International Conference on Electronic Technology， Communication and Information. Piscataway： IEEE， 2024： 1277-1280.
[9]	叶继华，时淑霞，李汉曦，等. 一种结合深度学习和随机森林的地平线检测方法［J］. 系统仿真学报， 2018， 30（7）： 2507-2514.
	YE J H， SHI S X， LI H X， et al. A horizon detection method based on deep learning and random forest［J］. Journal of System Simulation， 2018， 30（7）： 2507-2514.
[10]	LIU Y B， ZENG M， MENG Q H. Unstructured road vanishing point detection using convolutional neural networks and heatmap regression［J］. IEEE Transactions on Instrumentation and Measurement， 2021， 70： No.5002708.
[11]	ZHOU Y， QI H， HUANG J， et al. NeurVPS： neural vanishing point scanning via conic convolution［C］// Proceedings of the 33rd International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2019： 866-875.
[12]	陈星，张文海，杨林，等. 结合立体视觉的非结构化道路消失点检测研究［J］. 西北工业大学学报， 2022， 40（6）： 1431-1439.
	CHEN X， ZHANG W H， YANG L， et al. Research on vanishing point detection of unstructured road scene combined with stereo vision［J］. Journal of Northwestern Polytechnical University， 2022， 40（6）： 1431-1439.
[13]	LIU S， ZHOU Y， ZHAO Y. VaPiD： a rapid vanishing point detector via learned optimizers［C］// Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2021： 12839-12848.
[14]	KOCUR V， FTÁČNIK M. Traffic camera calibration via vehicle vanishing point detection［C］// Proceedings of the 2021 International Conference on Artificial Neural Networks， LNCS 12895. Cham： Springer， 2021： 628-639.
[15]	王伟，赵春辉，唐心瑶，等. 自适应地平线约束下的车辆三维检测［J］. 计算机应用， 2024， 44（3）： 909-915.
	WANG W， ZHAO C H， TANG X Y， et al. 3D vehicle detection with adaptive horizon line constraints［J］. Journal of Computer Applications， 2024， 44（3）： 909-915.
[16]	LIN T Y， GOYAL P， GIRSHICK R， et al. Focal loss for dense object detection［C］// Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway： IEEE， 2017： 2999-3007.
[17]	ZHANG X， LIU C， YANG D， et al. RFAConv： innovating spatial attention and standard convolutional operation［EB/OL］. ［2025-04-10］..
[18]	LIU W， LU H， FU H， et al. Learning to upsample by learning to sample［C］// Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2023： 6004-6014.
[19]	SOCHOR J， ŠPAŇHEL J， HEROUT A. BoxCars： improving fine-grained recognition of vehicles using 3-D bounding boxes in traffic surveillance［J］. IEEE Transactions on Intelligent Transportation Systems， 2019， 20（1）： 97-108.
[20]	DUBSKÁ M， HEROUT A. Real projective plane mapping for detection of orthogonal vanishing points［C］// Proceedings of the 2013 British Machine Vision Conference. Durham： BMVA Press， 2013： No.90.
[21]	KHANAM R， HUSSAIN M. YOLOv11： an overview of the key architectural enhancements ［EB/OL］. ［2025-06-16］. .
[22]	CARION N， MASSA F， SYNNAEVE G， et al. End-to-end object detection with Transformers［C］// Proceedings of the 2020 European Conference on Computer Vision， LNCS 12346. Cham： Springer， 2020： 213-229.
[23]	YANG L， TANG T， LI J， et al. BEVHeight++： toward robust visual centric 3D object detection［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2025， 47（6）： 5094-5111.
	王伟，张朝阳，唐心瑶，等. 道路场景下相机自动标定及优化算法［J］. 计算机辅助设计与图形学学报， 2019， 31（11）： 1955-1962.
	WANG W， ZHANG C Y， TANG X Y， et al. Automatic self-calibration and optimization algorithm of traffic camera in road scene［J］. Journal of Computer-Aided Design and Computer Graphics， 2019， 31（11）： 1955-1962.
[24]	王伟，唐心瑶，张朝阳，等. 跨相机交通场景下的车辆空间定位方法［J］. 计算机辅助设计与图形学学报， 2021， 33（6）： 873-882.
	WANG W， TANG X Y， ZHANG C Y， et al. Spatial positioning method of vehicle in cross-camera traffic scene［J］. Journal of Computer-Aided Design and Computer Graphics， 2021， 33（6）： 873-882.
[25]	SOCHOR J， JURÁNEK R， ŠPAŇHEL J， et al. Comprehensive data set for automatic single camera visual speed measurement［J］. IEEE Transactions on Intelligent Transportation Systems， 2019， 20（5）： 1633-1643.
[26]	HAO R， FAN S， DAI Y， et al. RCooper： a real-world large-scale dataset for roadside cooperative perception［C］// Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2024： 22347-22357.
[27]	FU J， ZHAO J， LI F， et al. Bisection window and homogeneity principle-based local contrast measure for infrared sea-sky line detection［J］. IEEE Geoscience and Remote Sensing Letters， 2024， 21： No.7000105.

面向跨相机鸟瞰视角道路对齐的地平线检测方法

Horizon detection method for cross-camera bird’s-eye view road alignment

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 27

相关文章 2

编辑推荐

Metrics

[1]	徐国愚, 闫晓龙, 张一丹. 基于动态上采样的轻量级生成对抗网络DU-FastGAN[J]. 《计算机应用》唯一官方网站, 2025, 45(10): 3067-3073.
[2]	王伟, 赵春辉, 唐心瑶, 席刘钢. 自适应地平线约束下的车辆三维检测[J]. 《计算机应用》唯一官方网站, 2024, 44(3): 909-915.