indiradutta/Pose-Estimation

Pose Detection

This repository holds the tensorflow implementation of Single Person Pose Detection using Movenet and Apache MXNet implementation of Multi-Person Pose Detection using Openvino in Python.

Dataset

The model 'Movenet' is trained on MS COCO dataset.
The model 'ssd_512_mobilenet1.0_coco' which is an object detection framework under OpenVino is trained on VOC2007, VOC2012 and MS COCO dataset.

Steps involved

Single Person Pose Detection

• Imported the necessary libraries as mentioned in the 'Requirements' section.
• Defined the 'Movenet' model.
• Defined the function for drawing the landmarks and initialized the edges.
• Ran inference for pose detection using opencv.

Multi-Person Pose Detection

• Imported the necessary libraries as mentioned in the 'Requirements' section.
• Defined the 'ssd_512_mobilenet1.0_coco' model for object detection.
• Modified the detector model such that it is made to detect only 'human beings'.
• Defined the 'simple_pose_resnet18_v1b' model for pose detection.
• Ran inference for pose detection using opencv.

Requirements

numpy==1.21.0
opencv-python==4.5.3
tensorflow==2.5.0
matplotlib==3.4.2
future==0.18.2
mxnet==1.8.0
gluoncv=0.10.4

---

Demo Video

Single Person Pose Detection

https://www.youtube.com/watch?v=pVcm9asUgZg

Multi-Person Pose Detection