T2FPV-ETH
收藏资源简介:
T2FPV-ETH数据集是由卡内基梅隆大学创建的,旨在支持第一人称视角(FPV)轨迹预测研究。该数据集通过模拟真实世界中的顶视图轨迹数据集,生成了所有交互行人的自我中心视觉数据。T2FPV-ETH数据集特别关注由于不完美的检测和跟踪、遮挡以及相机视野限制(FOV)引起的FPV特定错误。数据集的创建过程涉及使用SEANavBench高保真模拟环境,提供真实的合成图像,并提供每个轨迹的历史和未来对应的地面实况。该数据集的应用领域主要集中在解决机器人导航中的FPV错误问题,提高预测性能,减少位移误差。
The T2FPV-ETH dataset was developed by Carnegie Mellon University to support research on first-person view (FPV) trajectory prediction. This dataset generates egocentric visual data for all interacting pedestrians by simulating real-world bird's-eye view trajectory datasets. The T2FPV-ETH dataset specifically focuses on FPV-specific errors caused by imperfect detection and tracking, occlusions, and camera field-of-view (FOV) constraints. The dataset was constructed using the SEANavBench high-fidelity simulation environment, which provides photorealistic synthetic images, and includes corresponding ground truth for both historical and future segments of each trajectory. Its core application scenarios focus on resolving FPV-related errors in robot navigation, enhancing prediction performance, and reducing displacement error.

- 1T2FPV: Dataset and Method for Correcting First-Person View Errors in Pedestrian Trajectory Prediction卡内基梅隆大学 · 2023年



