NHL arenas use a combination of sensor-embedded pucks and infrared cameras mounted in the rafters to track the puck in real time. The puck itself contains tiny LEDs that beam infrared light outward, and anywhere from 16 to 28 specialized cameras hung overhead detect that light and triangulate the puck’s position on the ice. This system, called NHL Edge, runs alongside the broadcast cameras you see on TV to generate the tracking data behind replays, on-screen graphics, and live stats.
What’s Inside the Puck
Every NHL game puck now has sensors and infrared LEDs built directly into it by SMT, the league’s tracking technology partner. The LEDs sit near the top and bottom surfaces of the puck, and their light escapes through six tiny tubes called “light pipes” that channel infrared light outward in a wide cone shape. This design ensures the light reaches overhead cameras no matter how the puck is oriented, whether it’s flat on the ice, flipping through the air, or banked off the boards.
SMT’s sensors are integrated during manufacturing by Inglasco, the company that has supplied NHL pucks since 1996. The re-engineered design moved the LEDs closer to the puck’s surface, which disperses infrared light much more broadly than earlier versions that sent light out in narrow beams. According to SMT’s CEO Gerard Hall, this “bright star” puck gives the tracking system far better coverage and more precise location data.
The Camera Network Overhead
Each NHL arena has between 16 and 28 infrared-sensitive cameras installed in the rafters, positioned to cover every angle of the ice surface. These aren’t the broadcast cameras you see on the TV feed. They’re dedicated optical tracking cameras whose only job is picking up the infrared light from the puck (and from sensors stitched into player jerseys). By combining signals from multiple cameras at different angles, the system can triangulate the puck’s position in three dimensions, including its height off the ice.
All of this data flows through SMT’s software platform, called Oasis, which processes the raw camera feeds and distributes real-time tracking information. That data powers everything from the speed-of-shot graphics you see during broadcasts to the advanced stats available through NHL Edge.
Why Tracking a Puck Is So Difficult
A hockey puck is one of the hardest objects in professional sports to track visually. In a standard broadcast frame, the puck occupies roughly 0.005% of the total pixels. It’s a small, dark, fast-moving disc against a white surface, and it regularly disappears behind players’ bodies, gets lost in motion blur during slap shots, or blends into dark patches like the goal line or rink borders.
Computer vision researchers have documented a long list of challenges. Player skates and broadcast overlays (score bugs, network logos) create visual artifacts that algorithms can mistake for the puck. When the puck sits on a painted line, the contrast drops and it becomes nearly invisible. During scrums along the boards or in front of the net, multiple players can completely block the camera’s view for several seconds at a time. This is precisely why the NHL moved to an active infrared system rather than relying purely on what cameras can “see” in visible light.
How Software Fills the Gaps
Even with infrared sensors, the puck’s signal can momentarily drop out. When that happens, software algorithms estimate where the puck is based on its last known trajectory and speed. More advanced approaches use contextual cues from the players themselves. Researchers have developed deep learning models that analyze player positions, body poses, and formations to predict where the puck likely is, even when it’s completely hidden from view. The logic is straightforward: if five players are converging on a specific spot along the boards, the puck is probably there too.
One recent approach segments each player from the background and feeds those positions into a neural network alongside the video frame. This lets the system combine what it can see directly (the puck’s infrared signal or its visible appearance) with what it can infer from context. The model learns patterns like how players orient their sticks and bodies relative to the puck, which helps it maintain tracking through the chaotic moments that pure visual detection would miss entirely.
The Original Glow Puck
The concept of putting technology inside a hockey puck dates back to 1996, when Fox Sports introduced FoxTrax, better known as the “glow puck.” That system used 20 infrared emitting diodes packed inside the puck, arranged in five strings of four: four on top, four on the bottom, and twelve around the perimeter. Infrared cameras attached to computers around the arena detected the light, while optical encoders on the broadcast camera tripods reported each camera’s pan and tilt angle. An analog-to-digital converter tracked each camera’s zoom level. All of this data fed into a system that overlaid a blue glow (or a red comet tail for fast shots) onto the broadcast feed.
FoxTrax was polarizing with fans, and Fox dropped it after a few seasons. But the core idea, embedding infrared emitters in the puck and detecting them with specialized cameras, is essentially the same principle the NHL uses today. The difference is that modern hardware is smaller, more reliable, and paired with vastly more powerful software. Where FoxTrax was a broadcast novelty, NHL Edge is a full data platform generating hundreds of data points per second for every player and the puck simultaneously.