基于稀疏WiFi信号与视觉图像融合的室内定位方法研究

With the increasing demands on location-based services in indoor environments, indoor positioning has become one of the most attractive research fields. Regarding as promising techniques for indoor localization, pure vision-based and WiFi-based solutions have their own advantages and disadvantages. In this research project, we will propose indoor positioning techniques based on the fusion of sparse WiFi signals and visual images to get over these limitations and obtain better performance overall. Firstly, we will investigate the image-driven representation of wireless localization which establishes a wireless sensing model with high precision and embeds all received sparse signals onto a single image. Secondly, we will explore a multi-modal dictionary learning based localization algorithm with the fusion of wireless signals and visual images, it is implemented by performing pedestrian detection using improved visual background extractor and combining the detection results with the wireless localization images to locate persons. Finally, we will further investigate the fusion localization methods under dynamic indoor environments. The transfer learning will be adopted to reduce the labeling time and manual effort required, also a method will be proposed to reduce the localization errors caused by environmental changes based on online learning techniques. This research project aims to make a theoretical and technical breakthrough in positioning based on fusion of wireless and visual data, and provide new solutions for real-time location sensing in indoor environments.

随着基于位置的服务需求日益增长，室内定位技术备受国内外研究人员的关注。WiFi定位和视觉定位作为近年来新兴定位技术，具有各自的优势与局限性。本项目拟研究基于稀疏WiFi信号和视觉图像融合的定位技术，弥补单一定位方式的不足，实现更好的定位性能。首先研究非视距环境下基于稀疏无线信号的图像化定位方法，建立准确可靠的无线信号感知模型，使用图像分析显示无线信号。其次，研究基于多模态字典学习的无线信号与视觉图像融合定位方法，采用优化视觉背景提取算法检测室内行人，将检测结果与无线信号图像化结果相结合估计室内行人位置信息。最后，进一步研究动态室内环境下的融合定位方法，使用迁移学习节省人工标注样本的开销，利用在线学习降低动态环境因素导致的定位误差。本项目的目标是实现无线信号与视觉图像融合定位的理论与技术突破，为实时室内位置感知提供新的解决方案。

室内用户信息智能感知在老年人健康护理、智能家居、灾难救援等领域具有广阔的应用前景，备受国内外研究人员的关注。WiFi感知和视觉感知作为近年来新兴感知技术，具有各自的优势与局限性。本项目研究了基于稀疏WiFi信号和视觉图像融合的感知技术，弥补单一感知方式的不足，实现更好的感知性能。首先研究了基于稀疏WiFi信号的室内非视距环境感知技术，建立了准确可靠的无线信号感知模型。其次，研究了基于多模态融合的无线信号与视觉图像感知技术，采用多模态深度融合无线信号不同模态信息及无线信号与视觉图像信息，实现准确的室内感知模型。最后，研究了动态室内环境下的感知技术，使用半监督学习节省人工标注样本的开销，利用对抗学习等技术完成感知模型在动态环境下的精确感知。本项目的研究成果可应用于室内环境下的用户智能感知，以推进人机协助等领域技术的实际应用。