---
title: "Sit-Up Counter With Pose Estimation: Landmarks, Camera, Rep Logic"
canonical: "https://aifitnessapi.com/motion/sit-up-counter-pose-estimation"
cluster: "AI Motion"
primary_query: "sit up counter pose estimation"
last_reviewed: "2026-10-18"
description: "Counting sit-ups from a camera, per the studies: elbow-to-knee distance and body angle, why the camera view decides recognition, rep logic and limits."
publisher: "AIFitnessAPI — independent, not sponsored"
cite_as: "\"Sit-Up Counter With Pose Estimation: Landmarks, Camera, Rep Logic\", AIFitnessAPI, https://aifitnessapi.com/motion/sit-up-counter-pose-estimation"
---

# Sit-Up Counter With Pose Estimation: Landmarks, Camera, Rep Logic

> A camera sit-up counter in one 2025 study used two signals: the distance between elbow and knee, which falls below a threshold when the person sits up, and the angle between the body and the horizontal, which falls below another threshold when they lie back down. Camera angle decides whether a pose model can see the movement at all: in a 2026 study, a camera branch recognised sit-up frames 95.0% of the time from the side, 88.5% from the front and 25.7% from behind. The 2025 study describes its thresholds as adjustable, so calibrate them for your camera, model and users.

- Canonical: https://aifitnessapi.com/motion/sit-up-counter-pose-estimation
- Last reviewed: 2026-10-18
- Publisher: AIFitnessAPI (https://aifitnessapi.com) — independent, not sponsored
- Cite as: "Sit-Up Counter With Pose Estimation: Landmarks, Camera, Rep Logic", AIFitnessAPI, https://aifitnessapi.com/motion/sit-up-counter-pose-estimation

---

This page is the sit-up layer under [how rep counting works](/motion/how-rep-counting-works) and [how form feedback works](/motion/how-form-feedback-works): landmarks, camera placement, rep logic and what the two studies we read on October 18, 2026 report. Each number is one study's result under that study's conditions, so use it as a starting point to calibrate for your camera, model and users. This is engineering guidance for building a tracker, not exercise or medical advice.

## The studies behind this page

- **Lei and colleagues** (Sensors, 2025) built a fixed-camera assessment system on [MediaPipe](/motion/mediapipe-pose-landmarker-models)'s BlazePose landmarks with state machines for pull-ups, push-ups and sit-ups. Their sit-up result is a demonstration with one test-taker rather than an accuracy study.
- **Gu and colleagues** (Sensors, 2026) recorded 10 participants, 6 men and 4 women, doing 30 repetitions each of push-ups, sit-ups and squats from the side, the front and behind, and report how often their camera branch recognised each exercise from each view. They fused the camera with a nine-axis inertial sensor; this page uses only the camera results.

## Landmarks and angles

Lei and colleagues define the top of the sit-up by distance and the bottom by angle: when leaning up, the distance between the elbow and the knee is less than a sit-up threshold, and when lying down, the angle between the body and the horizontal line is less than a sit-down threshold. Gu and colleagues kept 17 of MediaPipe's 33 keypoints, the shoulders, elbows, wrists, hips, knees and ankles plus head and foot reference points, and expressed them relative to the hip centre to reduce the influence of body size and camera distance.

Every keypoint scheme on this site has elbows, knees, shoulders and hips, from [COCO 17](/pose-keypoints/coco-17) to [BlazePose 33](/pose-keypoints/blazepose-33). In our reading, an elbow-to-knee distance in pixels changes with the person's size and distance from the camera, so normalise it, for example by torso length, before you compare it with a threshold.

## Where to put the camera

Gu and colleagues' camera branch recognised sit-up frames 95.0% of the time from the side, 88.5% from the front and 25.7% from behind. From behind, 675 of 1000 sit-up frames went unrecognised and 68 more were assigned to the wrong exercise; the authors attribute the posterior decline, for push-ups and sit-ups in particular, to core keypoints occluded by the trunk or the ground.

Lei and colleagues dedicate one camera to each person and recommend one camera per person when several people are tested at once. They add that their calibration-free 2D design is potentially vulnerable to large changes in viewing perspective or camera position, so in our reading a side camera, fixed once, is the set-up to start from.

## A rep state machine for sit-ups

Lei and colleagues' sit-up machine has a sitting-up state, a lying-down state, transition states between them and a misaligned state, and they say its transitions follow the same principle as their pull-up machine, where each counted repetition is one pass from State4 through State3 to State1. Lei and colleagues describe their thresholds as empirical values set from assessment standards or from each individual's performance, and say the sit-up threshold can be adjusted for actual conditions. In our reading, using a distance at the top and an angle at the bottom gives the two-threshold hysteresis described in [how rep counting works](/motion/how-rep-counting-works) without having to tune one signal for both ends.

The same system cleans keypoints before the state machine sees them, correcting sudden jumps they call flying points and smoothing over a 10-frame window. Gu and colleagues gate the camera on confidence instead: when the mean keypoint confidence in a window falls below 0.45, the visual contribution is switched off, a value they describe as a fixed engineering setting rather than an optimised one.

For form feedback, Gu and colleagues list partial sit-up completion as an example of insufficient movement amplitude, which in our reading is a repetition that leaves the lying-down state but never satisfies the sitting-up threshold.

## Failure modes the studies report

- **A camera behind the person.** Gu and colleagues' posterior view lost most sit-up frames, and the only posterior frames assigned to the wrong exercise, 68 of them, all came from the sit-up subset.
- **Occlusion and poor light.** Lei and colleagues saw keypoints jump when occlusion occurred or lighting was poor, which is what their correction step targets.
- **A moved camera.** Lei and colleagues' calibration-free design is potentially vulnerable to large changes in viewing perspective or camera position.

[Pose estimation accuracy](/motion/pose-estimation-accuracy) covers the general causes in more depth.

## What was measured, and why it will not transfer

- Gu and colleagues' frame-level figures measure whether the camera branch recognised the exercise, not whether repetitions were counted. Across the three exercises, the micro-averaged accuracy was 98.1% from the side, 89.4% from the front and 30.6% from behind.
- Lei and colleagues report counting accuracy for push-ups and pull-ups but, in the text we read, not for sit-ups; their sit-up experiment concludes that one test-taker performed 7 standard sit-ups.

No study we read reports sit-up counting accuracy across a group of users, so treat a sit-up counter as unvalidated until you have tested it yourself; [testing a rep counting algorithm](/test/rep-counting) covers how.

## Sources

- [pmc.ncbi.nlm.nih.gov/articles/PMC12610833/](https://pmc.ncbi.nlm.nih.gov/articles/PMC12610833/) — Lei et al., Sensors 2025: sit-up states and thresholds, keypoint correction, camera specification and the sit-up demonstration (read as Europe PMC's XML). (checked 2026-10-18)
- [pmc.ncbi.nlm.nih.gov/articles/PMC13611031/](https://pmc.ncbi.nlm.nih.gov/articles/PMC13611031/) — Gu et al., Sensors 2026: sit-up recognition by camera view, keypoint selection, confidence gate and feedback labels (read as NCBI's efetch XML). (checked 2026-10-18)

## FAQ

### Which landmarks does a sit-up counter use?

Lei and colleagues used the distance between elbow and knee for the sitting-up position and the angle between the body and the horizontal for the lying-down position. Every common keypoint scheme has elbows, knees, shoulders and hips, so the signals are available from any model; normalise the distance before you threshold it.

[Permalink](https://aifitnessapi.com/motion/sit-up-counter-pose-estimation#faq-1)

### Where should the camera go to count sit-ups?

To the side, on the evidence we read: Gu and colleagues' camera branch recognised sit-up frames 95.0% of the time from the side, 88.5% from the front and 25.7% from behind, where the trunk and the ground hid core keypoints. Keep the camera fixed, since Lei and colleagues note their calibration-free design is vulnerable to large camera moves.

[Permalink](https://aifitnessapi.com/motion/sit-up-counter-pose-estimation#faq-2)

### How accurate is camera-based sit-up counting?

No study we read reports it across a group of users. Gu and colleagues measured exercise recognition rather than counting, and Lei and colleagues demonstrated their sit-up counter on one test-taker, so measure your own counter before relying on it.

[Permalink](https://aifitnessapi.com/motion/sit-up-counter-pose-estimation#faq-3)