| 研究課題/領域番号 |
19K24364
|
|---|
| 研究機関 | 大阪大学 |
|---|
研究代表者 |
|
|---|
| 研究期間 (年度) |
2019-08-30 – 2021-03-31
|
| キーワード | Multi-Gait Modeling / Dense Trajectories / Feature Representation |
|---|
| 研究実績の概要 |
The main goal of this project is to recognize the identities of the persons while they are walking in groups. Due to the lack of sufficient multi-gait datasets that fit in my task, I currently trying to benefiting from the existing data in my lab trying to adapt it to fit in my project objectives.The current data I compiled contains both visual data and sensor data. The visual data consists of the single person walking sequences and the multi persons walking sequences. As well, the sensor data contains the time stamped readings of the attached sensor on the person torso during his/her walking. Over all more than 150 subject data are collected.
Achievements for visual data:A-Create Single person reference model: for this goal I did;
a.Extract the foreground pixels for from all single person walking sequences.
b.Extract the dense trajectories and optical flow information from single gait sequences. B-For multi-person walking sequences (a).Match the multi person sequences with the corresponding single person’s sequences.(b).I applied the recent Multi-objects tracking and segmentation technique to segment and track each person within the walking group separately.
Achievements for sensor data (1)-I am working synchronizing the timestamps of both the image sequences and sensor data. As the sensor data come into one time-stamped sequence which mix the single walking with the group walking sequence. Therefore. (2) I have visualized the data trying to separate the single sensor data to create the reference model.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Due to the lack of multi-gait dataset, I spent some time to create my own dataset the is already compiled from the existing data in my lab.
|
| 今後の研究の推進方策 |
1-Use both the extracted visual and sensor data of the single walking sequences to build a reference model for each subject.
2-Extract the bounding boxes for each person within walking group.
3-Extract the dense trajectories for the walking group sequences and associate them to each corresponding bounding box.
4-Propose a feature representation method to create global feature descriptor for each walking person in both single and group walking sequences.
5-Once the global feature descriptor for a subject within the group is created, I can measure the pairwise similarity between the single gait reference model and the corresponding extracted descriptor from multi-gait sequences.
|
| 次年度使用額が生じた理由 |
for support publication and travel expenses relevant to the project and the coming months
|
