Future Person Localization in First-Person Videos

[jayakornv]

Paper: http://openaccess.thecvf.com/content_cvpr_2018/papers/Yagi_Future_Person_Localization_CVPR_2018_paper.pdf

Short summary: Predict the future location of a pedestrian in egocentric camera

Details

• Method : Supervised deep neural networks
• Input: from previous T frames,
  • Pose of target prdestrain
  • Location + scale of the pedestrain
  • Egomotion info (learned visual odometry info)
• Output : Predicted location of the pedestrian in the next t frames.

Github: https://github.com/takumayagi/fpl

[Fujiry0]

This paper does not consider multiple possible trajectories, interpersonal relations (social norms) and image semantics.