VIP-Mobility360: From RNIB Voices to Real-World Data – A Comprehensive Video Dataset of Outdoor Mobility Obstacles for People with Visual Impairments

While variety of object detection datasets exist, few address the specific and practical challenges faced by blind and visually impaired persons (VIPs) in real-world outdoor mobility contexts. Existing datasets typically focus on general-purpose objects or indoor scenes and overlook dynamic and environmental factors that critically affect safer mobility in daily life. To address this gap, we introduce VIP-Mobility360 — a novel, primary annotated dataset uniquely designed to support AI models that enhance outdoor mobility for VIPs. VIP-Mobility360 is the first dataset curated in direct response to real-world obstacle reports collected based on Royal National Institute of Blind People (RNIB) reports, consultations, and global case studies involving hazards such as e-scooters, bollards, wheelie bins, construction barriers, and pavement cracks. The dataset contains high-resolution video recordings (100 sequences per object class) captured entirely in outdoor environments, incorporating diverse weather conditions, background variations, scale, orientation, and 360° camera perspectives. This design ensures data realism, robustness, and applicability to real-time tracking, geometric understanding, and time-series forecasting. All instances are manually annotated to support supervised learning, object detection, depth estimation, and semantic segmentation. Beyond assistive technology, VIP-Mobility360 has broader relevance in areas such as autonomous vehicles (e.g., puddle and hazard detection), generative AI for geometric reconstruction, and urban planning for inclusive design. Moreover, by advancing the development of AI-driven smart canes, wearable mobility aids, and context-aware guidance systems, this dataset contributes to reducing reliance on costly personal carers or guide dogs—offering a socially equitable and economically viable solution to independent living for the blind community. This dataset therefore represents a transformative step toward building trustworthy, inclusive, and ethically aligned AI applications in real-world healthcare and urban mobility contexts.

尽管现有各类目标检测数据集层出不穷，但鲜有数据集能够针对盲人与视障人士（Visually Impaired Persons, VIPs）在真实户外出行场景中面临的具体实用挑战进行适配。现有数据集通常仅聚焦于通用目标或室内场景，却忽略了在日常生活中对安全出行至关重要的动态因素与环境变量。为填补这一研究空白，我们推出了VIP-Mobility360——一款全新的原始标注数据集，专为赋能支持视障群体户外出行的人工智能模型而设计。VIP-Mobility360是首款基于英国皇家盲人协会（Royal National Institute of Blind People, RNIB）的报告、咨询调研以及全球案例研究中收集的真实障碍物场景数据所构建的数据集，所涵盖的障碍物包括电动滑板车、路桩、带轮垃圾桶、施工围挡以及路面裂缝等。该数据集包含全户外场景下采集的高分辨率视频片段（每类目标对应100个序列），涵盖多样化的天气条件、背景差异、目标尺度、拍摄角度以及360°相机视角。该数据集的设计方案保障了数据的真实性、鲁棒性，同时使其能够适配实时跟踪、几何理解以及时序预测等任务。所有目标实例均经过人工标注，可用于监督学习、目标检测、深度估计以及语义分割等任务的模型训练。除辅助技术领域外，VIP-Mobility360在其他多个领域也具有广泛应用价值，例如自动驾驶汽车（如水坑与障碍物检测）、用于几何重建的生成式AI（Generative AI）以及包容性设计导向的城市规划等。此外，该数据集能够推动智能盲杖、可穿戴出行辅助设备以及情境感知导航系统等AI驱动技术的发展，有助于减少视障群体对昂贵的私人陪护或导盲犬的依赖，为盲人群体的独立生活提供兼具社会公平性与经济可行性的解决方案。因此，该数据集为在真实医疗保健与城市出行场景中构建可信、包容且符合伦理规范的人工智能应用迈出了变革性的一步。

搜集汇总

背景与挑战

背景概述

VIP-Mobility360是一个专门针对视障人士户外移动障碍物的视频数据集，基于皇家盲人协会（RNIB）报告和全球案例研究收集真实世界障碍物（如电动滑板车、护栏等），包含高分辨率视频序列（每类100个），涵盖多样户外环境、天气条件和360度视角，适用于对象检测、深度估计等AI任务，旨在提升辅助技术和城市规划的包容性。