PanoPed：超越边界框的全景行人仿真到真实跟踪
原标题：PanoPed: Beyond Bounding Boxes for Sim-to-Real Panoramic Pedestrian Tracking
AI 摘要
该论文提出PanoPed，一个面向全球面行人跟踪的仿真到真实基准，包含108,000帧合成数据（固定、四足机器人、无人机相机）和28,002帧真实数据（其中16,247帧密集标注），每帧同步提供掩码、深度、相机位姿和3D行人状态。作者指出ERP边界框无法唯一确定行人在球面上的中心与角范围，并提出Sextant——一个仅约0.035M参数、即插即用的角度定位头，复用冻结检测器且不改变跟踪身份。在PanoPed-S测试中，Sextant将最强基线MOTIP的HOTA从47.30提升至49.49，且无需真实数据微调即可在真实视频上提升MOTIP和HAT 0.96–1.14 HOTA。
正文节选
PanoPed: Beyond Bounding Boxes for Sim-to-Real Panoramic Pedestrian Tracking Abstract Full-sphere panoramic cameras let fixed monitoring systems and mobile robots track people in every direction, but a planar bounding box does not fully describe where a person is on the sphere. We introduce PanoPed, a sim-to-real benchmark for pedestrian tracking on the full sphere. PanoPed-S contains 108,000 frames from fixed, quadruped-mounted, and drone-mounted cameras, with synchronized masks, depth, camera