基于位置-迭代双重感知的双目相机外参在线微调方法

doi:10.16182/j.issn1004731x.joss.26-0414

摘要/Abstract

摘要：

双目系统在长期运行中，温度、振动冲击等因素导致外参发生微小漂移，漂移累积使测距误差随距离放大，严重损害远距测距能力。由于在线补偿困难，这些微小漂移常被忽略。为此，提出一种基于位置-迭代双重感知的在线微调方法。建立乘法耦合权重仿真模型，融合空间可靠度与迭代阶段调制函数；设计三阶段渐进优化算法，平衡收敛速度与鲁棒性；构建面向离线标定与在线微调的仿真系统，建立面向高可信初值的快速收敛机制。在EuRoC MAV数据集上进行验证，所提方法的归一化残余误差为0.601，绝对旋转误差较离线标定初值显著降低，平均迭代18.9次，运行时间约23.5 ms，修正精度与收敛稳定性均优于对比方法。蒙特卡罗仿真验证了算法鲁棒性，系统集成仿真展示了可行性，KITTI Odometry跨场景验证进一步表明了算法在真实道路环境中的基础泛化能力。所提方法为双目相机外参在线微调提供了有效建模与优化方案。

关键词: 双目相机标定, 外参微调, 位置-迭代感知, 渐进优化, 蒙特卡罗仿真

Abstract:

In long-term operation of binocular systems, small drifts in extrinsic parameters may occur due to temperature changes, vibration, and mechanical shocks. As these drifts accumulate, the ranging error increases with distance and seriously degrades long-range measurement performance. Because online compensation is difficult, such small drifts are often neglected. To address this problem, an online fine-tuning method based on position-iteration dual perception is proposed. A multiplicative coupling weighting model is established by integrating spatial reliability with an iteration-stage modulation function. A three-stage progressive optimization algorithm is designed to balance convergence speed and robustness. A simulation framework for offline calibration and online fine-tuning is constructed, together with a fast convergence mechanism for high-confidence initial values. Experiments are conducted on the EuRoC MAV dataset, the proposed method achieves a normalized residual error of 0.601, with the absolute rotation error significantly reduced compared with the offline calibration initialization. The method converges in an average of 18.9 iterations with a runtime of approximately 23.5 ms per frame, demonstrating higher correction accuracy and more stable convergence than the compared methods. Monte Carlo simulations verify the robustness of the algorithm, and system-level simulations demonstrate its feasibility. Cross-scenario validation on the KITTI Odometry dataset further indicates the algorithm's basic generalization ability in real road environments.The proposed method provides an effective modeling and optimization scheme for online fine-tuning of binocular camera extrinsic parameters.

Key words: binocular camera calibration, extrinsic fine-tuning, position-iteration dual perception, progressive optimization, Monte Carlo simulation

中图分类号:

TP391.9

龙潜,迪力萨提·吐尔洪,郑淇等 . 基于位置-迭代双重感知的双目相机外参在线微调方法[J]. 系统仿真学报, 2026, 38(9): 2437-2449.

Long Qian,Dilixiati·Tuerhong,Zheng Qi,et al . Online Fine-tuning of Binocular Camera Extrinsic Parameters Based on Position-iteration Dual Perception[J]. Journal of System Simulation, 2026, 38(9): 2437-2449.

图/表 13

图1

图2

表1

三阶段渐进微调算法

阶段	迭代范围	$c s p a (p)$	收敛条件
I保守微调	$[0, K 1)$	0.6	$‖ Δ R ‖ < 0.01 °$
II中等调整	$[K 1, K 2)$	0.6→0.3	$‖ Δ R ‖ < 0.005 °$
III精细优化	$[K 2, K m a x]$	0	$m a x i \| r i (I t e r) - r i (I t e r - 1) \| < 10 - 4$

表1

图3

表2

实验中扰动、噪声与异常值的生成方式

类型	具体描述与参数
在线微调扰动	对双目相机的真实外参施加微小随机扰动，以验证算法的自恢复能力。旋转扰动：随机轴+随机旋转角 $Δ θ ~ U (- 0.5 °,$ 0.5°)，平移扰动： $Δ t ~ U (- 1 %, 1 %)$ 相对于基线
特征点高斯噪声	引入蒙特卡罗噪声，模拟图像传感器噪声的标准差 $σ : 0.5,1.0,2.0$ 像素
异常值比例	模拟错误匹配的特征点比例为 $0,5 %, 10 %, 20 %$

表2

表3

表4

在线微调场景仿真结果

方法	$R e$ /(°)	$ρ$	迭代次数
SpatialOnly	0.229±0.121	1.027	28.5±7.8
IterativeOnly	0.195±0.108	0.874	25.7±6.9
Additive	0.162±0.095	0.726	22.4±5.7
本文	0.134±0.082	0.601	18.9±4.5

表4

表5

场景泛化结果

场景	$R e /$ (°)	$ρ$
easy	0.134	0.601
difficult	0.262	1.175

表5

表6

消融实验结果

变体	组件	$R e /$ (°)	$ρ$
完整算法		0.134±0.082	0.601
变体1	移除空间加权	0.195±0.108	0.874
变体2	移除迭代调制	0.229±0.121	1.027
变体3	乘法→加法	0.162±0.095	0.726
变体4	仅中心特征	0.149±0.088	0.668

表6

表7

蒙特卡罗结果

噪声/像素	异常值比例	$R e$ /(°)	$ρ$	成功率/%
0.5	0	0.134±0.082	0.601	98
0.5	10	0.162±0.095	0.726	92
1.0	0	0.229±0.121	1.027	88
1.0	10	0.305±0.158	1.368	82
2.0	0	0.462±0.215	2.072	75
2.0	10	0.669±0.284	3.000	68

表7

表8

表9

时序滤波消融实验

方法	$R e$ /(°)	$ρ$
单帧(无时序)	0.134	0.601
SMA (窗3)	0.120	0.350
SMA (窗5)	0.197	0.370
EMA ( $α$ =0.3)	0.110	0.305
EMA ( $α$ =0.5)	0.114	0.315
卡尔曼滤波	0.181	0.342

表9

表10

固定参数和自适应参数对比

方法	$R e /$ (°)	$ρ$
固定参数	0.134	0.601
自适应参数	0.127	0.432

表10

参考文献 23

[1]	王若萱, 吴建平, 徐辉. 自动驾驶汽车感知系统仿真的研究及应用综述[J]. 系统仿真学报, 2022, 34(12): 2507-2521.
	Wang Ruoxuan, Wu Jianping, Xu Hui. Overview of Research and Application on Autonomous Vehicle Oriented Perception System Simulation[J]. Journal of System Simulation, 2022, 34(12): 2507-2521.
[2]	Wang Kunfeng, Zhao Kaichun, Lu Wenshuai, et al. Temperature Drift and Compensation in Stereo Visual Odometry[J]. IEEE Sensors Journal, 2024, 24(23): 39744-39750.
[3]	He Zhicheng, Tan Juntao, Lin Zhigui, et al. Online Extrinsic Parameters Calibration of On-board Stereo Cameras Based on Certifiable Optimization[J]. Measurement, 2025, 242: 115911.
[4]	全国汽车标准化技术委员会. 智能网联汽车组合驾驶辅助系统技术要求及试验方法第1部分:单车道行驶控制: [S]. 北京: 中国标准出版社, 2024.
	National Automotive Standardization Technical Committee. Intelligent and Connected Vehicle—Technical Requirements and Testing Methods for Combined Driver Assistance System—Part 1: Single-lane Manoeuvre: [S]. Beijing: Standards Press of China, 2024.
[5]	Ling Yonggen, Shen Shaojie. High-precision Online Markerless Stereo Extrinsic Calibration[C]//2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway: IEEE, 2021: 1771-1778.
[6]	Dexheimer E, Peluse P, Chen Jianhui, et al. Information-theoretic Online Multi-camera Extrinsic Calibration[J]. IEEE Robotics and Automation Letters, 2022, 7(2): 4757-4764.
[7]	Kanai T, Vasiljevic I, Guizilini V, et al. Robust Self-supervised Extrinsic Self-calibration[C]//2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway: IEEE, 2023: 1932-1939.
[8]	Chen Yanshun, Yan Weiqing, Yue Guanghui, et al. Multi-resolution Hybrid Explicit Representation for Novel View Synthesis of Dynamic Scenes[J]. IEEE Transactions on Intelligent Vehicles, 2025, 10(2): 1117-1127.
[9]	Zhang Z. A Flexible New Technique for Camera Calibration[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(11): 1330-1334.
[10]	Heikkila J, Silven O. A Four-step Camera Calibration Procedure with Implicit Image Correction[C]//Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 1997: 1106-1112.
[11]	Abad Francisco, Camahort Emilio, Vivó Roberto. Camera Calibration Using Two Concentric Circles[C]//Image Analysis and Recognition. Berlin, Heidelberg: Springer Berlin Heidelberg, 2004: 688-696.
[12]	Sturm Peter F, Ramalingam S, Tardif J P, et al. Camera Models and Fundamental Concepts Used in Geometric Computer Vision[J]. Foundations and Trends in Computer Graphics and Vision, 2011, 6(1/2): 1-183.
[13]	Chen Siyu, Ma Chao, Liu Chao, et al. Feature Selection Based on the Self-calibration of Binocular Camera Extrinsic Parameters[J]. International Journal of Wavelets, Multiresolution and Information Processing, 2024, 22(1): 2350030.
[14]	Lepetit Vincent, Moreno-Noguer Francesc, Fua Pascal. EPnP: An Accurate O(n) Solution to the PnP Problem[J]. International Journal of Computer Vision, 2009, 81(2): 155-166.
[15]	Kneip Laurent, Li Hongdong, Seo Yongduek. UPnP: An Optimal O(n) Solution to the Absolute Pose Problem with Universal Applicability[C]//Computer Vision-ECCV 2014. Cham: Springer International Publishing, 2014: 127-142.
[16]	Li Shiqi, Xu Chi, Xie Ming. A Robust O(n) Solution to the Perspective-n-point Problem[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(7): 1444-1450.
[17]	姜丽梅, 陈信威. 动态场景下基于特征点筛选的视觉SLAM算法[J]. 系统仿真学报, 2025, 37(3): 753-762.
	Jiang Limei, Chen Xinwei. Visual SLAM Algorithm Based on Feature Point Selection in Dynamic Scenes[J]. Journal of System Simulation, 2025, 37(3): 753-762.
[18]	Yi Hao, Liu Bo, Zhao Bin, et al. Outliers Rejection for Robust Camera Pose Estimation Using Graduated Non‐convexity[J]. IET Computer Vision, 2025, 19(1): e12330.
[19]	Triggs Bill, McLauchlan P F, Hartley R I, et al. Bundle Adjustment—A Modern Synthesis[C]//Vision Algorithms: Theory and Practice. Berlin, Heidelberg: Springer Berlin Heidelberg, 2000: 298-372.
[20]	Yuhai Oleksandr, Cho Yubin, Choi Ahnryul, et al. Enhanced Three-axis Frame and Wand-based Multi-camera Calibration Method Using Adaptive Iteratively Reweighted Least Squares and Comprehensive Error Integration[J]. Photonics, 2024, 11(9): 867.
[21]	Xie Qiwei, Long Qian, Li Jianping, et al. Application of Intelligence Binocular Vision Sensor: Mobility Solutions for Automotive Perception System[J]. IEEE Sensors Journal, 2024, 24(5): 5578-5592.
[22]	Zhao Liangtian, Xu Xiangmin, Pei Shanshan, et al. Road Surface State Change Detection Based on Binocular Vision for Autonomous Driving System[J]. ACM Transactions on Autonomous and Adaptive Systems, 2025, 20(4): 29.
[23]	Xie Qiwei, Hu Xiyuan, Ren Lei, et al. A Binocular Vision Application in IoT: Realtime Trustworthy Road Condition Detection System in Passable Area[J]. IEEE Transactions on Industrial Informatics, 2023, 19(1): 973-983.

方法名称	核心特点	与本文方法的关键区别
StandardLM	标准LM优化，所有观测点单位权重	缺乏对异常值和空间异质性的感知
SpatialOnly	仅基于特征点空间位置分配权重	忽略了优化迭代过程中的残差动态变化和迭代阶段的鲁棒性需求
IterativeOnly	仅根据优化迭代进程动态调整权重	忽略了特征点固有的空间可靠性差异，可能导致低质量边缘特征被错误赋高权重
Additive	空间权重与迭代权重采用加法组合	无法实现对异常值和不可靠观测点的有效抑制
本文	基于位置-迭代双重感知，乘法耦合权重	通过乘法耦合融合空间可靠度与迭代调制，有效抑制异常。三阶段优化兼顾收敛鲁棒，实现在线微调快速收敛

模块	时间/ms	占比/%
特征匹配	12.5	53.2
空间可靠度计算	1.2	5.1
权重计算	0.8	3.4
LM迭代(平均18.9次)	9	38.3