Computer Vision Engineer — Pose Estimation + Sports Biomechanics Metrics (Video → JSON/CSV)

I’m building a sports “Scout Metrics” feature that turns short athlete videos into pose keypoints + biomechanics metrics + a quality score.

Goal (MVP):
Given a video clip (phone or fixed camera), the system outputs:
1. Pose keypoints per frame (MediaPipe 33 landmarks or equivalent) with confidence
2. A Quality Gate that returns PASS / WARN / FAIL with reason codes (framing, lighting, missing joints, confidence stability, occlusion, etc.)
3. Event/phase detection (start simple) and compute a set of biomechanics metrics
4. Exports: JSON + CSV (deterministic schema)

Sport focus to start: Quarterback throwing motion (football)
(If you’ve done baseball throwing, golf swing, tennis swing—mention it.)

Key requirements:
• Must produce deterministic, repeatable outputs (same clip → same results)
• Prefer Python pipeline (OpenCV + MediaPipe / ONNX / TensorRT okay)
• Must include confidence smoothing / temporal filtering
• Deliver a clean, documented repo + sample outputs for test clips

Deliverables:
• Working pipeline (video → keypoints → quality gate → metrics)
• Reason codes for WARN/FAIL
• Example outputs for 3–5 clips
• Clear documentation + how to run locally
• Optional: lightweight API wrapper (FastAPI)

Nice-to-have:
• Experience quantizing models / running on edge devices (Raspberry Pi / CM5)
• Sports motion analysis experience

To apply:
Please send:
1. 1–2 relevant examples (pose estimation / motion analysis / sports)
2. Your proposed approach for quality gate + event detection
3. Estimated timeline for MVP + cost
4. Tech stack you recommend (MediaPipe vs alternate model)