Computer vision rep counting is the automatic detection and counting of exercise repetitions using a smartphone camera and machine learning. The system tracks body landmarks in real time, identifies the start and end positions of a movement cycle, and increments a counter each time a full repetition is completed — all without wearable hardware. Unlike accelerometer-based rep counting in smartwatches, computer vision rep counting observes the full body, enabling simultaneous form quality assessment during every rep.
Computer vision rep counting (noun) — A method of automated exercise tracking that uses a smartphone camera and pose estimation models to detect and count repetitions in real time. The system identifies movement phase transitions (start position → peak contraction → return) and increments a rep counter at each completed cycle, enabling hands-free, hardware-free workout tracking with simultaneous form analysis.
Computer vision rep counting operates through a three-stage pipeline that runs in real time on the smartphone's neural processing unit:
The camera feed is processed frame-by-frame by a pose estimation model (such as Apple's Vision framework or MediaPipe). The model outputs 25–33 body landmark positions — joints including shoulders, elbows, wrists, hips, knees, and ankles — at each frame.
A phase detector analyzes the trajectory of key landmarks over time to identify the characteristic positions of each exercise: the starting position (e.g., arms extended for a curl), the peak contraction (e.g., arms fully flexed), and the return to start. Each exercise has a pre-defined phase model.
A state machine tracks transitions between phases. When the system detects a complete transition from start → peak → start, it increments the rep counter. Partial reps, pauses, and failed reps are detected separately and flagged in the session summary.
| Factor | Computer Vision | Wearable (Apple Watch, etc.) |
|---|---|---|
| Hardware needed | Smartphone only | Smartwatch ($200–$800) |
| Form analysis | Yes — full body | No — wrist motion only |
| Accuracy | 95–99% (good lighting) | 85–95% (exercise dependent) |
| Setup | Position camera, press record | Always on, automatic |
| Exercises supported | Any visible movement | Limited to wrist-dominant exercises |
| Real-time feedback | Yes — form + rep count | Rep count only |
SportsReflector's AR Live Training mode is the leading implementation of computer vision rep counting for consumer fitness. The feature uses ARKit and Apple's Vision framework for real-time body tracking, automatically counting repetitions while simultaneously displaying form feedback on screen. At the end of each set, athletes receive a session summary showing total reps, form score per rep, and the highest-priority corrections for the next set. The system supports all major gym exercises and sports drills, making it the only rep-counting solution that also functions as a full AI coaching platform.
Computer vision rep counting is the automatic detection and counting of exercise repetitions using a smartphone camera and machine learning. The system tracks body landmarks in real time, identifies the start and end positions of a movement cycle (for example, the bottom and top of a bicep curl), and increments a counter each time the full range of motion is completed. Unlike wearable-based rep counting (which uses accelerometers), computer vision rep counting requires no hardware beyond a smartphone and can simultaneously assess form quality during each repetition.
Computer vision rep counting works in three stages. First, pose estimation extracts body landmark positions from each video frame — typically 25–33 joints including shoulders, elbows, wrists, hips, knees, and ankles. Second, a movement phase detector identifies the key positions that define one repetition: the starting position, the peak contraction, and the return to start. Third, a state machine tracks transitions between these phases and increments the rep counter when a full cycle is completed. Advanced systems also detect partial reps, pauses, and failed reps, and can trigger real-time form feedback at the moment each rep is completed.
Computer vision rep counting works for any exercise with a clear, repeating movement cycle. Supported exercises include free weights (bicep curls, tricep extensions, shoulder press, lateral raises, dumbbell rows, bench press, dumbbell flyes), bodyweight exercises (push-ups, pull-ups, dips, squats, lunges, burpees, mountain climbers, jumping jacks), cable exercises (cable curls, tricep pushdowns, cable rows, face pulls), and barbell exercises (squats, deadlifts, bench press, overhead press, Romanian deadlifts). Sports drills with repetitive cycles — such as jump rope, box jumps, and kettlebell swings — are also supported.
Computer vision rep counting accuracy is typically 95–99% for controlled exercises in good lighting conditions. Accuracy decreases for exercises with ambiguous movement cycles (such as deadlifts, where the start and end positions are similar), fast movements that blur between frames, or environments with poor lighting or multiple people in frame. SportsReflector's AR Live Training mode uses computer vision rep counting with real-time form feedback, providing both an accurate rep count and technique scoring for each repetition.
Wearable rep counting uses accelerometers and gyroscopes in devices like Apple Watch or Fitbit to detect movement patterns and infer rep counts. It is convenient (always on the wrist) but less accurate for exercises where the wrist is not the primary moving segment, and it cannot assess form quality. Computer vision rep counting uses a camera to observe the full body, enabling both accurate rep detection and simultaneous form analysis. The trade-off is that computer vision requires a camera to be positioned facing the athlete, while wearables work automatically without setup.
SportsReflector uses computer vision rep counting in its AR Live Training mode. The feature provides automatic repetition detection for all supported exercises, real-time form feedback during each rep, and a session summary with total reps, form scores, and corrective recommendations. It is available in the Pro tier at $9.99/month on iOS.