Dataset

We used the TBD Pedestrian Dataset[1] by CMU. It has the following salient features:

Multiple static cameras (3 views) with intrinsic and extrinsic matrix provided for each camera.
Bird’s eye view scene (matching our target domain)
Provides ground human truth trajectory (3D points projected to Z=0 plane)
Each trajectory matched with pedestrian ID across all 3 viewpoints.
Provides Frame level trajectory coordinates

TBD Pedestrian Dataset setup – provides multi-view time-synchronized video feed with camera matrices

2D labeled trajectories for each pedestrian in all 3 camera views. Pedestrians are matched and given the same ID across all 3 views.

References

[1] Wang, A., Biswas, A., Admoni, H., & Steinfeld, A. (2022). Towards Rich, Portable, and Large-Scale Pedestrian Data Collection. ArXiv, abs/2203.01974