Pose Estimation AI

Train custom keypoints for people, parts, and equipment, and turn poses into step checks, guidance, and analytics.

Gloved hands assembling a metal component on a jig, overlaid with purple computer vision tracking lines, keypoints, and a text tag reading "step 3 · insert ok".

Pose Estimation AI for the Toughest Lines, Cells, and Fields of Play

  • Deploy Anywhere, Run Everywhere

    Run pose estimation at workstations, robot cells, and inspection stations, on the edge, on-prem, air-gapped, in your VPC, or via API.

  • One Platform, Full Adoption

    Tools every engineering team can adopt, from manufacturing and automation engineers to EHS leads and sports analysts, no separate ML team required.

  • Secure, Compliant, and Audit-Ready

    Data stays safe with SOC 2 Type II compliance, encrypted data, and an uptime SLA, with face blurring at the edge and keypoints stored instead of video where you choose.

  • Custom Keypoint Skeletons
  • Part & Equipment Keypoints
  • Step Verification from Body Pose
  • Sports Technique Analysis
  • Edge Real-Time Inference
  • Calibrated Per-Keypoint Confidence
“The vision AI solutions we developed with Roboflow have been a core component of connecting our teams and unlocking their ability to collaborate.”

Lou Stocco

Director of Manufacturing Advanced Analytics, USG

Talk to a Vision AI engineer who's shipped in manufacturing.

Bring us your toughest pose estimation problem and we'll map a working solution.

Ask us about:

  • Solution architecture that fits your skeletons, stations, and robot cells
  • A live demo on video and images from your own operations
  • Deployment options: edge, on-prem, air-gapped, or VPC
  • ROI modeling against rework, SOP misses, injuries, and engineering time

Over 80,000 organizations build with Roboflow.

  • Rivian
  • Pella
  • Chobani
  • USG Corporation
  • BNSF Railway
  • American Woodmark
Start where you are

Vision AI is transforming businesses

Customers across the board are solving complex challenges and driving meaningful impact.

  • Automotive customer

    $10 million

    Saved by automatically detecting defects on the production line

  • Logistics & freight company

    90%

    Less time spent manually tracking shipping inventory

  • Building materials supplier

    60%

    Lower customer return rate with improved product quality

Locate Every Keypoint, from Body Pose to Part Orientation

Add a keypoint layer to your vision applications with Roboflow Vision AI for pose estimation. Built for teams where a box around an object is not enough, where you need to know which way a part is facing, where a hand is on a fixture, or how a body moved, and where off-the-shelf pose models only know the 17 points of a human skeleton.

Train custom keypoints:

  • Define your own skeleton in Roboflow Annotate, with any number of points, on people, tools, parts, gauges, or equipment.
  • Fine-tune RF-DETR Keypoint on your own images, with code and weights under the Apache 2.0 license.
  • Get a calibrated confidence for every keypoint, so downstream logic knows which points to trust.

Industrial pose applications:

Deploy anywhere:

  • Run keypoint models in real time on an edge device at the station, or in the cloud on live streams.
  • Chain keypoints with detection, tracking, zones, and logic in Workflows to turn poses into counts, alerts, and pass or fail results.
  • Use the same models for sports analytics, from technique and rep counting to player tracking.

Bring intelligence to every pose today. Know where every point is, on people, parts, and machines.

‍

Frequently asked questions

What is pose estimation?

Pose estimation finds a set of keypoints on an object, such as the joints of a body or the corners and features of a part, and connects them into a skeleton. Keypoint detection is the general form of the same task, for any object. Read more about keypoint detection.

Can we train keypoints for tools, animals, or equipment, not just people?

Yes. You define the skeleton, so it can be 4 points on a valve handle, 20 points on a sports court, or the 17-point human body. RF-DETR Keypoint supports any number of keypoints and multiple keypointed classes in one model.

Which pose estimation models can we use?

Roboflow trains and deploys RF-DETR Keypoint, which leads the newest YOLO pose models on both accuracy and speed on COCO, along with other popular keypoint architectures. Compare the best pose estimation models.

Can it run in real time on the edge?

Yes. Keypoint models run on NVIDIA Jetson and other edge devices at the station, and send results over REST, MQTT, or a PLC connection.

Roboflow serves as the keypoint engine, and your engineering team owns the skeletons, step definitions, and pass or fail rules.

‍