HEADS-UP
收藏资源简介:
HEADS-UP是由瑞士洛桑联邦理工学院的VITA实验室创建的第一个头戴式第一人称视角数据集,专门用于盲人辅助系统中的轨迹预测。该数据集包含超过43,000帧的RGB图像、深度数据、点云数据、IMU测量值和行人轨迹标签,旨在解决盲人在动态环境中与行人碰撞的风险问题。数据集的创建过程包括使用ZED Mini立体相机进行数据采集,并通过视觉惯性里程计和YOLOv8模型进行数据标注。HEADS-UP数据集的应用领域主要集中在盲人导航系统的轨迹预测和碰撞检测,旨在通过预测行人的未来轨迹来提前预警潜在的碰撞风险。
HEADS-UP is the first head-mounted first-person view dataset developed by the VITA Lab at École Polytechnique Fédérale de Lausanne (EPFL), specifically designed for trajectory prediction in blind assistance systems. It contains over 43,000 frames of RGB images, depth data, point cloud data, IMU measurements, and pedestrian trajectory labels, targeting the mitigation of pedestrian collision risks for visually impaired individuals in dynamic environments. The dataset creation process involves data collection using a ZED Mini stereo camera, as well as data annotation via visual-inertial odometry and the YOLOv8 model. The main application scenarios of the HEADS-UP dataset focus on trajectory prediction and collision detection for blind navigation systems, with the objective of providing early warnings of potential collision risks by predicting pedestrians' future trajectories.



