Chris Pollett > Students >
Lodha

    ( Print View)

    [Bio]

    [Blog]

    [ Unity Hands-on ]

    [CS 297 Proposal]

    [Deliverable 1 ]

    [Deliverable 2]

    [Deliverable 3]

    [Deliverable 4]

    [Skeleton Based Action Recognition-PDF]

    [CS 297 Report-PDF]

    [CS298 Proposal]

    [Faster-RCNN-PDF]

    [CS298 Final Report-PDF]

    [CS298 Final Presentation-PDF]

CS 297 Deliverable 4

Human Bone Points Detection and Dataset Generation.

DESCRIPTION

This deliverable is aimed at identifying bone points of the humanoid avatar for the videos generated in Deliverable 3. Also a reseacrh paper on Skeleton based Action Recognition using CNN was studied.

The bone keypoints were detected using the OpenPose model.

About OpenPose : It is a realtime approach for single-person 3D keypoint detection : body, foot, hand, and facial keypoints.

About Skeleton based Action Recognition: This paper proposed a (CNN) based framework for both action classification and detection.

The video dataset generated here would be used to train an AI model to recognize the action performed by the humanoid avatar in a 3D space by implementing the approcach given in above paper.

PREVIEW

Input to the OpenPose Model

Input to the OpenPose model

Output from the OpenPose Model

Output to the OpenPose model

DOWNLOADS

Humanoid Skelton Video Dataset

Skeleton-based Action Recognition with Convolutional Neural Networks-PDF

REFERENCES

  1. Z. Cao, G. Hidalgo, T. Simon, S. Wei, and Y. Sheikh, "OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields," 2018. [Online]. Available: arXiv:1812.08008
  2. C. Li, Q. Zhong, D. Xie, and S. Pu, "Skeleton-based Action Recognition with Convolutional Neural Networks," 2017. [Online]. Available: arXiv:1704.07595