How PushLock Uses AI to Count Your Push-Ups

One of the most common questions about PushLock is: "How does it know I'm actually doing real push-ups?" The answer is on-device AI that runs entirely on your iPhone. fast, accurate, and completely private.

The Technology Behind Push-Up Detection

When you start a push-up session in PushLock, your iPhone's front camera activates. The app uses Apple's Vision framework to detect facial landmarks in real-time. specifically, the position of your face relative to the ground.

As you perform a push-up, your face moves in a predictable pattern: down toward the phone, then back up. PushLock tracks this movement frame by frame and counts a rep each time you complete a full down-up cycle.

How It Prevents Cheating

PushLock isn't just looking for any movement. The AI model validates several factors:

• Range of motion. your face needs to move a minimum distance to count as a rep. Tiny bobbing motions won't register.
• Consistent rhythm. the movement pattern must resemble actual push-up form, not random shaking.
• Face presence. the camera must see your face throughout the set. You can't just wave your hand in front of the phone.

The result is a system that reliably counts real push-ups while rejecting attempts to game it. Is it impossible to cheat? No. but it's hard enough that actually doing the push-ups is easier than trying to fake them.

Why On-Device Processing Matters

Every frame of camera data is processed entirely on your iPhone. Nothing is sent to a server. Nothing is stored. Nothing is uploaded. The camera feed is analyzed in real-time and immediately discarded.

This matters for two reasons:

Privacy

You're literally pointing a camera at your face while exercising. That's personal. PushLock was built with the principle that this data should never leave your device. The only thing saved is a number: how many push-ups you did.

Speed

On-device processing means zero latency. There's no waiting for a server response. Reps are counted instantly as you move, giving you real-time feedback during your set.

The Technical Details

For the technically curious, here's what's happening under the hood:

1. Camera capture. the front camera captures frames at 30fps using AVFoundation.
2. Face detection. Apple's Vision framework identifies facial landmarks (nose, eyes, chin) in each frame.
3. Position tracking. the vertical position of key landmarks is tracked over time, creating a movement signal.
4. Rep counting. the signal is analyzed for peaks and valleys. Each complete cycle (down then up) counts as one rep.
5. Validation. amplitude and timing thresholds ensure only legitimate reps are counted.

All of this runs on Apple's Neural Engine, which is specifically designed for machine learning workloads. It uses minimal battery and produces no noticeable heat during a typical set.

Accuracy in Real-World Conditions

PushLock's push-up detection works in a range of conditions:

• Lighting. works in normal indoor lighting. Very dark rooms may reduce accuracy.
• Distance. best results when your phone is on the ground 1-2 feet in front of you.
• Speed. handles both slow, controlled reps and fast-paced sets.
• Form variations. works with wide grip, narrow grip, and standard push-up positions.

What About Other Exercises?

Right now, PushLock focuses on push-ups. they're the perfect exercise for this use case because they require no equipment, can be done anywhere, and naturally position your phone to see your face. Push-ups are also a compound exercise that works your chest, arms, shoulders, and core all at once.

Try It Yourself

The best way to understand PushLock's AI push-up detection is to try it. Download the app, lock an app you use too much, and do your first set. You'll see each rep counted in real-time as you move. it's surprisingly satisfying to watch.