Pose Estimation 姿势估计

Pose Estimation is predicting the body part or joint positions of a person from an image or a video. 姿势估计预测图像或视频里人的身体部位或关节位置。

用人体关键点坐标以及连接来描述人的姿势和方位。

 

相关词包括：

posture analysis， 

 

可用的模型包括：

OpenPifPaf by VITA lab at EPFL.

TensorFlow pose estimation

https://www.tensorflow.org/lite/models/pose_estimation/overview

OpenPose

https://github.com/CMU-Perceptual-Computing-Lab/openpose

AlphaPose

DeepCut

Mask RCNN

https://arxiv.org/abs/1703.06870

3D human pose estimation in video with temporal convolutions and semi-supervised training

https://github.com/facebookresearch/VideoPose3D

 

 

Applications

1. Activity Recognition

Tracking the variations in the pose of a person over a period of time can also be used for activity, gesture and gait recognition. There are several use cases for the same, including:

• Applications to detect if a person has fallen down or is sick. 用于检测一个人跌倒或生病的应用程序。
• Applications that can autonomously teach proper work out regimes, sport techniques and dance activities. 可以自主教授适当的锻炼方式，运动技巧和舞蹈活动的应用程序。
• Applications that can understand full-body sign language. (Ex: Airport runway signals, traffic policemen signals, etc.). 可以理解身体手语的应用程序。 （例如：机场跑道信号，交警信号等）。
• Applications that can enhance security and surveillance. 可以增强安全性和监视性的应用程序。

2. Motion Capture and Augmented Reality

3. Training Robots

4. Motion Tracking for Consoles

5. Fitness coach apps

About AI fitness coach apps, the common flow looks as follows: 

1. Capture user’s movements while doing an exercise
2. Analyze the correctness of an exercise performance 
3. Display mistakes to the user interface

Digitalization has not spared the fitness industry. According to the Research and Markets report, the digital fitness market size is expected to reach $27.4 billion by 2022.

Assuming that the goal of the given system is to inspect the input video for common exercise mistakes and compare it with the reference video, where the professional athlete is performing the same exercise, the flow will look like as follows:

1. Cutting of the input video depending on the exercise start & end

2. Detecting 2D and 3D keypoints on the user’s body

3. Decomposing of the exercise phases

4. Searching for common mistakes

5. Comparing the input video frames with the reference ones

 

 

Literature 

https://towardsdatascience.com/human-pose-estimation-simplified-6cfd88542ab3

https://mobidev.biz/blog/human-pose-estimation-ai-personal-fitness-coach