【作者】 吴凡；

【导师】 赵方；

【作者基本信息】 北京邮电大学 ， 软件工程， 2025， 博士

【摘要】 在多源数据感知定位研究中,面向城区复杂环境的高精度定位算法是智能化城市建设和低空经济技术发展的核心,为自动驾驶、复杂环境救援、公共安全管理及城市智能交通提供了关键支撑。多源融合定位技术在解决复杂动态环境中的高精度、强鲁棒和快响应定位问题中发挥着至关重要的作用。开展复杂城区环境下的多源智能优化高精度定位研究,有助于政府提高城市交通、应急救援以及保障公共安全的能力,也能在日趋复杂的城区环境下为公众提供高精度、强鲁棒的定位服务。本文围绕复杂多变城区环境下的动态定位性能问题,结合GNSS定位优化技术、INS误差修正技术及自适应融合多源定位技术,设计了多体制导航平台优化策略,能够有效融合城市环境中的多源定位信息,提升定位精度和系统鲁棒性,确保高精度、强可靠定位结果。主要创新及贡献如下:(1)针对GNSS定位脆弱、受环境影响大问题,提出了一种基于D-Tran网络架构的GNSS定位优化算法。该算法充分利用Transformer的自注意力机制动态调整定位策略,通过整合历史轨迹信息与实时观测数据,有效提取复杂环境中的时空特征。同时,该方法还通过动态调整GNSS定位算法中的误差模型权重优化定位性能。实验结果表明,在多种复杂城区环境下,定位精度均优于最新的因子图优化GNSS定位算法和深度优化算法。(2)针对INS传感器的误差累积问题,提出了一种新颖的INS误差修正方法,并结合PSAF模型设计了误差估计与补偿方案。构建了 PSAF模型以精准捕捉INS传感器的漂移特性和动态误差规律,实现对误差的非线性建模与校正。实验结果表明,该方法降低了 INS误差的累积效应,有效减少了传感器误差对导航精度的影响。相较于最新的INS误差修正优化定位方法,所提出的方法INS定位精度在SPAN-CPT传感器数据集和ADIS16465传感器数据集上均有性能提升。(3)为实现多传感器数据的有效融合,提出了一种结合动态权重调整的因子图优化算法。该算法通过评估各传感器数据的质量,动态调整传感器在因子图优化定位中的权重分配,自适应优化多源传感器数据的融合结果,从而提升了系统对动态环境的适应能力和对传感器性能波动的鲁棒性,保证在各种环境条件下始终维持较高的定位精度。实验结果表明,该算法与高斯及Huber权重调整策略相比均能带来有效提升。(4)开发了一套面向复杂城区环境的多源融合定位验证系统,该系统集成了本文所提的优化算法,支持GNSS、INS、视觉及几何约束信息等多源传感器数据融合分析。通过该系统,实现了复杂场景下的高效计算与实时可视化功能,根据不同的导航需求和环境变化自动调整工作模式。系统通过对多种传感器信息的综合利用,提升了定位的灵活性和精确性,能够有效应对高楼遮挡、信号多径等挑战,为多源融合定位技术的实际应用提供了有力的支撑。本文通过结合GNSS定位优化、INS误差修正、多源系统自适应融合以及多源融合定位验证系统,提出了一种创新的城市复杂环境高精度定位优化方法。多层次、多角度解决了传统定位技术在城市环境中的多径效应、信号遮挡和误差累积等挑战。通过理论创新与实践验证相结合,本文为提升复杂场景下的定位精度和鲁棒性提供了重要理论依据和技术支持。该方法在智能交通、自动驾驶和机器人导航等领域具有广泛的应用前景,尤其适用于高动态环境和信号受限条件下的高精度定位需求。更多还原

【Abstract】 In the research of multi-source data perception positioning algorithms,high-precision positioning algorithms for complex urban environments are at the core of intelligent urban construction and low-altitude economic technology development,providing key support for autonomous driving,complex environment rescue,public security management,and urban intelligent transportation.Multi-source fusion positioning technology plays a crucial role in solving the problems of high-precision,strong robustness,and fast response positioning in complex dynamic environments.Conducting research on multi-source intelligent optimization for high-precision localization in complex urban environments helps governments enhance their capabilities in urban traffic management,emergency rescue,and public safety assurance.Moreover,it enables the public to access highly accurate and robust localization services even in increasingly complex urban settings.This study focuses on the challenges of dynamic localization performance in complex and dynamic urban environments.By integrating INS error correction techniques,GNSS positioning optimization strategies,and adaptive multi-source fusion localization methods,a multi-system navigation platform optimization strategy is designed.A multi-source fusion localization validation system is established to effectively integrate multi-source localization information in urban environments,improving localization accuracy and system robustness,ensuring highly precise and reliable localization results.The main innovations and contributions are as follows:(1)To address the vulnerability of GNSS positioning and its high susceptibility to environmental factors,this study proposes a GNSS positioning optimization algorithm based on the D-Tran network architecture.This algorithm leverages the self-attention mechanism of Transformers to dynamically adjust the positioning strategy.By integrating historical trajectory data with real-time observations,it effectively extracts spatiotemporal features in complex environments.Additionally,the method optimizes positioning performance by dynamically adjusting the weight of error models in the GNSS localization algorithm.The experimental results show that on positioning datasets of urban areas with different levels of complexity,the positioning accuracy is superior to the latest factor graph optimization GNSS positioning algorithm and the deep optimization algorithm.(2)To mitigate the issue of error accumulation in INS sensors,this study proposes an innovative INS error correction method and incorporates a PSAF model to design an error estimation and compensation scheme.The PSAF model is constructed to accurately capture the drift characteristics and dynamic error patterns of INS sensors,enabling nonlinear modeling and real-time correction of errors.Experimental results demonstrate that this method significantly reduces the cumulative effect of INS errors,effectively minimizing the impact of sensor errors on navigation accuracy.Compared with the latest INS error correction and optimization positioning method,the proposed method has improved positioning accuracy of INS on both the SPAN-CPT sensor dataset and the ADIS 16465 sensor dataset.(3)To enable effective multi-sensor data fusion,this study proposes a factor graph optimization algorithm incorporating dynamic weight adjustment.The algorithm evaluates the quality of each sensor’s data and dynamically adjusts the weight distribution of sensors within the factor graph optimization framework.This adaptive optimization enhances the fusion results of multi-source sensor data,significantly improving the system’s adaptability to dynamic environments and its robustness to sensor performance variations,ensuring high localization accuracy across different environmental conditions.The experimental results show that this algorithm can bring effective improvements compared with the Gaussian and Huber weight adjustment strategies.(4)A multi-source fusion localization validation system tailored for complex environments was developed in this study.The system integrates the proposed optimization algorithms,supporting real-time fusion and analysis of multi-source sensor data,including GNSS,INS,vision,and geometric constraint information.Through this system,efficient computation and real-time visualization of complex scenarios are achieved.and it can automatically adjust its operational mode based on different navigation requirements and environmental changes.By comprehensively utilizing multiple sensor information sources,the system significantly enhances localization flexibility and accuracy.It effectively addresses challenges such as urban high-rise occlusions and signal multipath effects,providing strong support for the practical application of multi-source fusion localization technology.By combining GNSS positioning optimization,INS error correction,adaptive multi-source system fusion,and a multi-source fusion localization validation system,this study proposes an innovative high-precision localization optimization method for complex urban environments.This method comprehensively addresses the limitations of traditional localization techniques,including multipath effects,signal obstructions,and error accumulation in urban environments.Through a combination of theoretical innovations and practical validations,this research provides a solid theoretical foundation and technical support for improving localization accuracy and robustness in complex scenarios.The proposed method has broad application prospects in intelligent transportation,autonomous driving,and robotic navigation,making it particularly suitable for high-dynamic environments and signal-constrained conditions requiring high-precision localization.更多还原