MBPR: Multi_Body_Pose_Reconstruction

English | 简体中文

Paper：《Real-Time Reconstruction of Multi-Body Pedestrian Pre-Impact Posture in Collision Accidents From Monocular Images》

1.Shoulders Of Giants：

[1] SMPL：A parameterized human body model

[2] SPIN：3D human reconstruction method

[3] human_model_viewer：Visual adjustment of SMPL parameters

[4] SMPL_Tools：This is based on human_model_viewer, we developed a target SMPLannotation tool

2.Pipline

3.Tutorial

0. Download：git clone https://github.com/wmj142326/MBPR

1. Environment configuration:SPIN/README.md

2. Runestimation.py

   python estimation.py --checkpoint=data/model_checkpoint.pt --img_file=input_img/01 --outfile=out/01

   input_img/01 is a folder to store input images

   out/01 is a folder to store the output results，include：pic_outfile, pic_outfile_shape, pkl_outfile

3. copy folders：

   madymo/images == input_img/01;

   madymo/pkl_file == out/pkl_outfile;

4. Run ped_pkl2xml.py：

   python ped_pkl2xml.py

5. The output is saved in a folder：madymo/xml_file/

6. Visualization：Import madymo/xml_file/*.xmlinto MADYMO software for viewing.

4.Appendix

In madymo/mesh/ folder，we provided the SMPL .obj results, Visualization by SMPL_Tools， need.pkl to .ini：

python ped_pkl2ini.py

5. Citation

@ARTICLE{10746249,
author={Wang, MeiJun and Meng, Yu and Xu, Yan and Li, Quan and Nie, Bingbing},
journal={IEEE Transactions on Intelligent Transportation Systems}, 
title={Real-Time Reconstruction of Multi-Body Pedestrian Pre-Impact Posture in Collision Accidents From Monocular Images}, 
year={2025},
volume={26},
number={1},
pages={457-471},
keywords={Pedestrians;Accidents;Injuries;Image reconstruction;Videos;Three-dimensional displays;Pose estimation;Computational modeling;Shape;Real-time systems;Pre-impact posture;pose reconstruction;multi-body pedestrian model;SMPL model;reconstruction of pedestrian-vehicle collision},
doi={10.1109/TITS.2024.3486214}}