patent.fyi
Applications  /  US 2026/0355574
United States Patent Application Publication

US 2026/0355574 A1

Published
Nov 12, 2026
Filed
Jan 7, 2026
Earliest date on the face
May 12, 2025
Classification
A63B 24/0006
First published here
Nov 19, 2026

Generated automatically. Not reviewed by a person. Not legal advice or an opinion.

Section 1 · machine-generated

What if?

Questions about how individual parts of claim 1 might be changed. A question may describe something different from what the claim language recites, or a version of it; no view is offered on which.

1.1External camera with live preview

What ifWhat if a tripod-mounted external camera supplies the live exercise video?
Claim language“a front camera of the user device”

The external camera sends 30 frames per second over a wired connection to the user device. The device displays the incoming frames as a live preview.

1.2Pose inference on sampled frames

What ifWhat if the system runs pose inference on every third video frame?
Claim language“analyses each frame of the captured live video”

A 60-frame-per-second video stream supplies every third frame to the pose model, yielding 20 landmark measurements per second. A temporal filter interpolates landmark positions between measurements for the preview overlay.

1.4Workout bounds from floor markers

What ifWhat if the system sets the workout box from floor markers before the user enters the camera view?
Claim language“anchoring the created virtual workout box to the initial user pose”

Four markers with known spacing establish the floor plane and locate the corners of the exercise area. The system projects a volume above that area into the image and retains the resulting bounds for the exercise.

1.5Continuous analysis with position flags

What ifWhat if movement analysis continues as the user crosses the workout box boundary?
Claim language“pauses and resumes data analysis based on user position”

The system calculates joint angles for all visible joints and tags each measurement with its position relative to the workout box. The repetition counter credits a repetition when its start and finish both fall inside the box.

1.7Rolling evaluation during each movement

What ifWhat if the system evaluates deviations in a rolling time window during the movement?
Claim language“the accumulated deviations are evaluated at repetition completion boundaries”

The system compares joint angles with phase-specific reference ranges and reviews the latest 250 milliseconds of deviations at each update. It issues a correction when an angle remains beyond its reference range for at least 200 milliseconds.

Section 2 · machine-generated

Variations

Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.

Other field of use

Movement feedback for warehouse lift training

A front-facing camera at a warehouse training station captures a trainee lifting an empty practice crate between a floor mark and a waist-height shelf. The software confirms full-body visibility from the setup stance and anchors a fixed tracking region around that stance. It counts a cycle when the trainee rises from the pickup position and returns upright after placing the crate. During each cycle it records trunk lean and left-to-right knee flexion differences, then displays a corrective cue when the measured motion departs from the configured training pattern.

Component substitution

Pose tracking with a depth camera

A front-facing time-of-flight depth camera supplies synchronized color and depth frames to a skeleton tracker that takes the place of the color-only pose estimator. The tracker returns three-dimensional joint coordinates and per-joint confidence values, and the interface projects visible joints onto the color preview. After confirming visibility from head to feet, the software anchors a fixed three-dimensional workout volume to the initial pose and pauses scoring when required joints leave it. The software computes joint angles from three-dimensional limb vectors and uses their changes to count repetitions and select corrective cues.

Parameter or scale variation

Two users in one camera view

A front-facing camera records two users at 1920 by 1080 pixels and 30 frames per second, with each user starting in a separate half of the image. The software locks a workout box for each user after 15 consecutive frames of full-body detection, adds a lateral margin equal to 20 percent of that user's shoulder width on each side, and limits each box to its assigned half. It maintains separate repetition and form records and pauses scoring for a user after 6 consecutive frames outside that user's box. Scoring resumes after 9 consecutive frames with the required joints inside the box and visible.

Combination with existing systems

Camera coaching with a force platform

A dual force platform beneath the user's feet sends time-stamped vertical force samples to the device that runs the camera-based squat tutor. The device aligns the force samples with the corresponding video frames using a common clock and anchors the workout box to the user's initial standing pose. For each completed squat, it uses knee-angle changes to mark the descent and ascent and calculates each foot's share of the total vertical impulse over the ascent. When the force split falls outside the configured band, the interface displays a weight-distribution cue alongside the camera-based form feedback.

Automated or data-driven variant

Automatic calibration of movement range

The tutor records five slow practice repetitions after establishing the fixed workout box and segments the repetitions using reversals in the primary joint angle. It retains repetitions with sufficient landmark confidence and endpoints inside the configured exercise bounds, then stores the median minimum and maximum angles as the reference range for that user and exercise. During subsequent repetitions, it compares the measured endpoints with the stored reference and accumulates deviations within each movement phase. At repetition completion it displays a range cue when an endpoint differs from the reference by more than the configured tolerance.

Automatic selection of corrective cues

After each repetition, the tutor ranks detected form errors by their deviation from configured joint-angle limits and by how long they persist. For the highest-ranked error, it selects a short display cue from a table indexed by error type and recent cue response. The tutor measures the same error on the next completed repetition and records the change in deviation for that cue. Once several responses are available for an error, it favors the cue associated with the largest median reduction for that user.

Specific version of a part

Landmark confidence and visibility check

The device prompts the user to center their body and turn to the requested orientation, then overlays landmarks with confidence scores of at least 0.80. It confirms full-body visibility when the required landmarks from head to ankles appear in five consecutive overlaid frames, then calculates joint angles from landmark triplets.

Angle thresholds for squat counting

The system identifies a squat from coordinated hip and knee flexion followed by extension in successive landmark measurements. It counts one repetition when the knee angle drops below 100 degrees and then returns above 160 degrees.

Performance grading with rolling statistics

The system groups performance into three levels using the proportion of repetitions within form tolerances, the variability of peak joint angles, and the proportion of repetitions within a target duration. It adjusts the frequency and detail of feedback using the grade from the latest eight repetitions.

Heart rate based energy tracking

A chest strap supplies heart rate to an energy expenditure model that uses body mass, resting heart rate, and exercise duration. The system subtracts the estimated expenditure from logged intake and adjusts the next session's energy target using completed workload and changes in exercise difficulty.

Section 3 · quoted from USPTO records

Other work

Passages reproduced word for word from two earlier-published U.S. patent documents, placed beside the language of claim 1. No characterization is offered. Each passage is marked A or B for the document it comes from; ¶ numbers are the paragraph numbers in the USPTO’s published full text of that document.

Document AUS 2015/0105881 A1
Published
Apr 16, 2015
Filed
Oct 14, 2014
Class
A63B 71/06
Document BUS 2025/0022317 A1
Published
Jan 16, 2025
Filed
Jul 12, 2024
Class
G06V 40/25
Claim 1
Document A
Document B
1.1prompting a user to activate a front camera of the user device to capture a live video of the user performing an exercise, and allowing the user to visualize the exercise performance on a graphical user interface of the user device as the user performs the exercise;
The exercise machine 110 includes a media hub 120 and a user interface 125. The media hub 120, in some cases, captures images and/or video of the user 105, such as images of the user 105 performing different movements, or poses, during an activity.
B · US 2025/0022317 A1 · ¶ 41
1.2analyzing the captured live video using a pose estimation model and a pose estimation algorithm of the DFTA to detect and extract key points of the user, wherein the DFTA analyses each frame of the captured live video;
Various downsampling or encoding layers 420 convert the feature map 410, resulting in two downsampled heatmaps, a BBox (bounding box) heatmap 430 and a Keypoints heatmap 435. The BBox heatmap 430 includes peaks that correspond to the center of each person in the image, and the Keypoints heatmap 435 includes channel-wise peaks to the center of each keypoint.
B · US 2025/0022317 A1 · ¶ 86
1.3verifying orientation and visibility conditions through guided interaction by providing directional prompts to position the user for optimal key point detection, overlaying the optimal key points onto the analyzed frames of the live video of the user upon reaching a predefined detection confidence, and thereafter verifying the full-body visibility of the user by executing a key point-based visibility check based on the overlaid optimal key points, and, upon successful verification, the DFTA calculates joint angles of the detected and extracted key points to monitor posture and exercise movement of the user;
1.4creating a virtual workout box around the user based on an initial pose detected in the live video and maintaining the created virtual workout box as a persistent spatial constraint during performance of the exercise by anchoring the created virtual workout box to the initial user pose, without recalculating the created virtual workout box boundaries on a per-frame basis, wherein the created virtual workout box defines a region of interest for exercise tracking, wherein the DFTA creates the virtual workout box according to user-specific parameters and exercise-specific configuration data, and maintains the created virtual workout box unchanged until completion of each exercise;
FIG. 6 depicts a model or framework 600 that includes an ROIAlign (Region of Interest Align) operation to extract a small feature map from the BBox heatmap 430. The framework 600 utilizes a ROIAlign operation 610 with the person bounding boxes (BBox heatmap 430) on the image feature map to create person-localized feature maps, which are provided to additional downsampling and Fully Connected+Softmax layers 620 to predict or output a pose or pose heatmap 630.
B · US 2025/0022317 A1 · ¶ 89
1.5determining whether the live video of the user is within the created virtual workout box, monitoring and analyzing the exercise movement of the user using the detected and extracted key points and the calculated joint angles, wherein the DFTA pauses and resumes data analysis based on user position relative to the created virtual workout box and the front camera of the user device;
Computer 102 may prompt the user to move into a certain region relative to the image capturing device 126 and/or relative to the infrared transceiver 128 so that the user is within frame and/or range. When properly positioned, system 100 may process movement of the user.
A · US 2015/0105881 A1 · ¶ 57
1.6tracking the exercise movement of the user using the detected and extracted key points and the calculated joint angles to identify type of exercise performed by the user and count number of repetitions, wherein the DFTA continuously analyzes motion data to ensure precise motion capture;
System 100 may process data from the video and/or sensors to determine a number of repetitions of an exercise that a user has performed or duration of an exercise, and may estimate a number of calories burned by the user based on the repetitions and/or duration information and the one or more known values, such as may be obtained from MET tables.
A · US 2015/0105881 A1 · ¶ 75
The system 158, in some cases, can employ a multi-dimensional neural network that is trainable by any input signal that encodes movement (e.g., optical flow, Inertial Motion Units (IMUs) with gyroscope and accelerometer data, 2D or 3D human body keypoints, and so on). Given an input of one or several of the signals, the neural network performs multi-task prediction for exercise recognition and repetition detection or counting.
B · US 2025/0022317 A1 · ¶ 192
1.7identifying deviations from a predefined exercise motion pattern, wherein the deviations detected during an active movement phase are accumulated and used to generate corrective feedback, and the accumulated deviations are evaluated at repetition completion boundaries using cumulative user movement data of the identified exercise to detect errors in the exercise movement of the user, wherein the DFTA identifies incorrect form arising from improper execution or limited range of motion due to physical constraints or body structure anomalies;
1.8providing a real-time corrective media feedback to the user, by the DFTA, based on the detected errors to assist the user in maintaining proper form;
1.9classifying the exercise performance of the user based on the tracked movements and the detected errors, wherein the DFTA grades user skill levels and adapts feedback based on repetition accuracy, consistency, and speed of the user; and
1.10estimating energy expenditure and dynamically maintaining a calorie balance based on user-specific metabolic characteristics and adaptive performance analytics, wherein the DFTA autonomously regulates energy targets in response to variations in exercise difficulty and user progression to achieve a personalized fitness outcome.
System 100 may prompt a user to perform one or more exercises, monitor user movement while performing the exercises, and provide the user with an energy expenditure estimate based on their movement. System 100 may analyze a user's form to determine if the user is making an exercise more or less difficult, and adjust the energy expenditure estimate accordingly.
A · US 2015/0105881 A1 · ¶ 53

An empty cell means only that no passage was reproduced there.

Statements in these documents
While most people appreciate the importance of physical fitness, many have difficulty finding the motivation required to maintain a regular exercise program.
A · US 2015/0105881 A1 · ¶ 2
Aspects of this disclosure relate to processing of data taken while a user performs an athletic activity to determine an estimate of energy expenditure such as, for example, an amount of calories burned.
A · US 2015/0105881 A1 · ¶ 6
The users may be utilizing a fitness machine (e.g., a treadmill, a stationary bike, a strength machine, a stationary rower, and so on), or may be moving through the world on a bicycle.
B · US 2025/0022317 A1 · ¶ 2
The users can also be performing other activities that do not include an associated machine, such as running, strength training, yoga, stretching, hiking, climbing, and so on.
B · US 2025/0022317 A1 · ¶ 3
Cite aspatent.fyi, US 2026/0355574 A1, claim 1, first published 2026-11-19T13:13:53-05:00, https://patent.fyi/US20260355574/, SHA-256 4deeceb1ed5fcf1fdbafe8345a34c17bed61ee60efbeef6066ec53843a5f6d45
First published
Permanent addresshttps://patent.fyi/US20260355574/
Content fingerprint (SHA-256)4deeceb1ed5fcf1fdbafe8345a34c17bed61ee60efbeef6066ec53843a5f6d45
Methodv2026-11-18.2 · automated, not reviewed by a person
Correctionscorrections@patent.fyi