2020 Fiscal Year Annual Research Report
Toward a Multi-Gait Analysis/Recognition System
| Project/Area Number |
19K24364
|
|---|
| Research Institution | Osaka University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2019-08-30 – 2021-03-31
|
| Keywords | Multi-Gait Modeling / Gait Energy Image / Feature Representation / Dense Trajectories / Fisher vector encoding |
|---|
| Outline of Annual Research Achievements |
This ultimate goal of this project is to build multi-gait recognition system. Due to the lack of mutli-gait dataset, I spent the most part of this project period working on dataset preparation and developing a robust gait feature representation. I mainly depended on the available video recordings in my lab for group walking and human behavior analysis in the wild. I compiled my dataset from these recordings as follow;1-I prepared 182 video for the walking subjects while they are walking individually (single gait dataset).
2-I compiled a group walking video for the same walking subjects while they are walking together freely in outdoor environment (multi-gait dataset). Around 80 subject out of 182 appeared in this multi-gait dataset.3-For the single gait dataset, I have extracted the binary silhouette sequences and Spatial-pyramid Optical flow. I have used the recent deep learning models for both silhouette and optical flow extraction. 4-For multi-gait dataset, I applied the recent deep learning model (FairMOT) of multi object tracking the extract the Boundingbox sequences for each subject within the group. Afterward, I extracted the binary silhouette and optical flow information for each bounding box sequence.
5-I used the dense trajectory combined with the masked optical flow to build feature descriptor for each trajectory. As well, I aggregated the relative position for each trajectory regrading the wlking subject bounding box. 6. I used the fisher vector encoding to build the global descriptor for both single and multi gait features. Compute pairwise similarity.
|
