Biomechanical Keypoint Annotation for Wearable Shoe Cameras
Multi-point leg and foot keypoint labeling on fisheye video frames from shoe-mounted cameras, delivered for Outer Motion's natural motion analysis.
- Client
- Outer Motion
- Engagement
- Feb – Aug 2026
- Labels
- Multi-point keypoints
- Hired
- Directly

Project Overview
Outer Motion (formerly wearM.AI) builds wearable shoe-tracking devices that capture how people naturally move in daily life, not in the lab. Their motion models learn from video frames recorded by cameras mounted on the shoe, looking up at the wearer's legs and feet.
SevvenLabs annotated these biomechanical video frames with multi-point keypoints across both legs and feet. We worked directly with Co-founder & CEO Dr. Huawei Wang and manager Simon Crete, M.S., through every revision until the dataset was complete.
Objective
The goal was a training dataset the client's motion models could rely on:
Precise multi-point labeling
Place consistent keypoints on joints and body-shape outlines across both legs and feet, frame after frame.
Real-world footage
Label reliably even when frames are dark, blurred, distorted by the fisheye lens or partly out of view.
Exactly to the client's spec
Match the client's keypoint definitions and refine them together through review rounds.
See the data and the result

Our Approach
We treated the labeling spec as a shared document with the client and built quality checks into every batch.
Labeling spec with the client
Agreed keypoint definitions, skeleton connections and rules for hard cases such as hidden or blurred points before scaling up.
Frame-by-frame keypoint annotation
Annotators placed every keypoint on each video frame, following the agreed point order and connections.
QC review on every batch
A dedicated quality-control reviewer checked point placement and consistency before each batch was delivered.
Revision rounds with the client
We responded quickly to feedback from Dr. Huawei Wang and Simon Crete and applied each change across the dataset.
Challenges & Solutions
Shoe-mounted footage is far harder to label than standard pose data.
High data volumes
The project covered a large number of video frames that all needed the same careful, multi-point labeling.
We ran a batch workflow with steady deliveries, so the client could review early batches while the rest were labeled.
Low-quality frames
Frames from a camera on a moving shoe are often dark, motion-blurred and bent by the fisheye lens, with legs partly out of frame.
We agreed clear rules with the client for uncertain or hidden points, and reviewers double-checked the hardest frames.
Complex multi-point body-shape labeling
Each frame needed many keypoints in a fixed order across both legs and feet, where small mistakes break the skeleton.
Annotators were trained on the client's spec, and QC checked point order and connections, not just positions.
Results
The client received the full dataset and, in their words, “a complete, highly accurate dataset.”
Complete dataset delivered
Every frame in scope was labeled and passed QC before delivery.
Accuracy confirmed by the client
Outer Motion described the delivered dataset as highly accurate.
A partner they recommend
The client praised our fast response to feedback and would happily recommend SevvenLabs.
What the client said
We had a great experience working with Sevvenlabs on our biomechanical video-frame annotation project at Outer Motion (formerly wearM.AI). Even with high data volumes, low-quality frames, and complex multi-point body-shape labeling for our wearable shoe-tracking devices, the team worked closely with Dr. Huawei Wang and M.S. Simon Crete through every revision. They were quick to respond to feedback and delivered a complete, highly accurate dataset. We’re grateful for their thoughtful collaboration and would happily recommend them!
Biomechanical video-frame keypoint annotation

Outer Motion (formerly wearM.AI)Dr. Huawei Wang, Co-founder & CEO · Simon Crete, M.S., ManagerHard footage, careful labels
Wearable motion data is messy by nature. Clear specs, QC on every batch and quick revision rounds turned it into training data Outer Motion's models can trust.
Want to build a similar AI solution?
Let’s consult with your engineering and product team to engineer reliable, production-grade ML models.
