紀要論文

ステレオカメラを用いる歩容認証のための歩行特徴抽出手法

A method to extract features for gait recognition using a stereo camera

jpn

24321192

https://doi.org/10.15002/00026283

法政大学大学院情報科学研究科

法政大学大学院紀要. 情報科学研究科編

18

1

6

2023-03-24

Version of Record

A new method using a novel puppet model was proposed to measure gait features. OpenPose-based stereo measurement, which acquires 3D coordinates of human joints, was severely affected by noise due to data fluctuations and/or their missing. To reduce the noise, in the previously method, we have adjusted the parallax so that limb lengthes matched those. This greatly reduced the noise but was not accurate enough to extract gait features. The method proposed here we used, a 3D puppet model with fixed length limbs. Three-dimensional rotation angles and shifts of the model were predicted so that their joint projections onto the both stereo pair image match their observations. Eight gait features (gait cycle, gait speed, arms and legs features: swing width, bend, and forward movement time) were calculated from the matched 3D model and these were successfully applied as feature vectors to identify the gait of the four persons.

Article

AA12746425