Skip to main content

Solution Description

Smart gym solution: the NeoMind dashboard showing member training analysis, equipment occupancy and traffic trends

NeoMind dashboard: member training analysis, equipment occupancy and traffic trends at a glance

Background​

Gym operations have long struggled with a few blind spots: what members trained and how many reps they did relies purely on memory, leaving coaches without quantified guidance; which machines are busiest and which zones sit idle is opaque to management; and outsourcing video analytics means footage leaves the premises, raising member privacy concerns.

Gyms don't lack cameras — they lack the intelligence that lets cameras "understand" training: knowing who the member is, which machine they're using, what movements they performed and how long they trained.

Solution Overview​

The idea: one camera + one platform, turning workouts into structured data.

Deploy a single CamThink NeoEyes NE503 edge AI camera (Hailo-15H, 20 TOPS NPU) overhead to cover a mid-size training area of 10–20 people. The camera runs four models concurrently on-device — person detection, 17-keypoint pose, face detection and face features — handling member identification and trajectory tracking; the gym-tracker extension on the NeoMind platform takes care of member library matching, equipment zone detection, rep counting and training report generation.

The entire analysis runs on-premises: data stays on the NeoMind host and never passes through third-party clouds. Members participate only as face feature vectors — no raw footage is stored — balancing intelligence with privacy compliance.

Workflow​

  1. Member identification — face features are matched against the member library to recognize returning members automatically; new faces are auto-enrolled, with details filled in later
  2. Trajectory tracking — ByteTrack multi-object tracking outputs stable IDs and foot-point coordinates, covering the whole floor with a single camera
  3. Zone & equipment analytics — foot points hit pre-defined equipment zones (with 3-second dwell debouncing that filters passers-by) to determine when equipment use starts and ends
  4. Movement recognition & counting — pose keypoints combined with a joint-angle state machine recognize and count exercises such as squats, bench presses, biceps curls and rows, judging whether the range of motion is sufficient
  5. Training reports & operations overview — per member: duration, equipment, movement details and historical trends; per gym: occupancy, equipment usage and traffic by time slot

Key Benefits​

  • Single-camera coverage: 10–20 people in a mid-size gym with one NE503 — no multi-camera handoff
  • Four models on-device: 20 TOPS NPU local inference with zero cloud dependency and low-latency output
  • Privacy first: only face feature vectors are stored, never raw footage; training data stays on the gym's local host
  • Cable-free retrofit: PoE single-cable power and data, IP67 rated for humid environments — install and go
  • Visual ROI editing: rectangle, polygon and lasso tools to define equipment zones; changes take effect immediately
  • Auto enrollment: unknown visitors are automatically enrolled as profiles for admins to complete later

Outcome​

  • Every workout automatically produces a detailed report (duration, equipment, movements and reps) — traceable, comparable, with training suggestions
  • Coaches plan programs based on data reports, shifting guidance from gut feeling to evidence
  • Occupancy, equipment utilization and time-slot traffic are visible in real time, informing scheduling and floor decisions
  • Recognition accuracy depends on lighting, crowd density and occlusion; use on-site measurement as the baseline

Learn More​