{"id":2479,"date":"2021-09-02T09:00:31","date_gmt":"2021-09-02T01:00:31","guid":{"rendered":"https:\/\/blog.ailia.ai\/uncategorized\/movenet-skeleton-detection-model-for-videos-with-intense-motion\/"},"modified":"2025-05-20T16:04:11","modified_gmt":"2025-05-20T08:04:11","slug":"movenet-skeleton-detection-model-for-videos-with-intense-motion","status":"publish","type":"post","link":"https:\/\/blog.ailia.ai\/en\/tips-en\/movenet-skeleton-detection-model-for-videos-with-intense-motion\/","title":{"rendered":"MoveNet : Pose Estimation for Video with Intense Motion"},"content":{"rendered":"\n

Overview<\/strong><\/h3>\n\n\n\n

MoveNet\u00a0<\/em>is a pose estimation model released by\u00a0Google\u00a0<\/em>on May 17, 2021. Compared to conventional pose estimation models, it improves the detection accuracy in videos with intense motion. It is ideal for live fitness and sports applications.<\/p>\n\n\n\n

\"\"
Source: https:\/\/blog.tensorflow.org\/2021\/05\/next-generation-pose-detection-with-movenet-and-tensorflowjs.html<\/a><\/figcaption><\/figure>\n\n\n\n

<\/a>Next-Generation Pose Detection with MoveNet and TensorFlow.js<\/a><\/p>\n\n\n\n

Architecture<\/strong><\/span><\/h3>\n\n\n\n

MoveNet<\/em> is able to detect 17 two-dimensional keypoints with high speed and high accuracy. There are two models available, Lighting <\/em>and Thunder.<\/em> The former can be used for applications that require speed and the latter for applications that require accuracy. Both Lightning <\/em>and Thunder <\/em>can run at 30FPS or higher on desktop PCs, laptops, and smartphones.<\/p>\n\n\n\n

The architecture is similar to CenterNet<\/em><\/a>. The feature extractor is based on MobileNetV2 <\/em>to which Feature Pyramid Network<\/em> (FPN) was added. By setting output stride to 4, it can handle high resolution feature map output.<\/p>\n\n\n\n

\"\"
Source: https:\/\/blog.tensorflow.org\/2021\/05\/next-generation-pose-detection-with-movenet-and-tensorflowjs.html<\/a><\/figcaption><\/figure>\n\n\n\n

The output of the AI model is a person center heatmap, a keypoint regression field, a person keypoint heatmap, and a 2D per-keypoint offset field.<\/p>\n\n\n\n

\"\"
Source: https:\/\/blog.tensorflow.org\/2021\/05\/next-generation-pose-detection-with-movenet-and-tensorflowjs.html<\/a><\/figcaption><\/figure>\n\n\n\n

The model was trained using the COCO dataset and another Google\u2019s internal dataset called\u00a0Active<\/em>. One limitation of the COCO dataset is that it does not include data from harsh environments where poses change drastically or motion blur is present, making it unsuitable for fitness and dance apps. However Google\u2019s internal dataset is made of annotated yoga, fitness, and dance videos from YouTube. Only three frames are taken from each video to ensure diversity in the dataset.<\/p>\n\n\n\n

\"\"
Source: https:\/\/blog.tensorflow.org\/2021\/05\/next-generation-pose-detection-with-movenet-and-tensorflowjs.html<\/a><\/figcaption><\/figure>\n\n\n\n

Usage<\/strong><\/h3>\n\n\n\n

You can use MoveNet with ailia SDK on the video stream of a web camera with the following command.<\/p>\n\n\n\n

$ python3 movenet.py -v 0<\/code><\/p>\n\n\n\n

And here is the result you can expect.<\/p>\n\n\n\n

\n