Canon put a live basketball game on a remote 3D tabletop. The production pipeline is the real story.

Three locations instead of one production room

Canon says about 100 cameras captured basketball at Aichi International Arena, with 3D reconstruction performed at its Kawasaki office and viewing at a separate Tokyo site. Around 60 participants used mixed-reality devices to see the live action on a tabletop from selectable angles. The system combined Canon volumetric video and MREAL with NTT-led IOWN All-Photonics Network links.

The important reported change is that the reconstruction server did not have to sit at the arena. This was a demonstration, not a launch announcement for a consumer service. Lumacta did not attend or test it.

Source notes: 1. Analysis and proposed examples are identified in the text.

This is not simply choosing another broadcast camera

Canon’s technology explanation describes capturing a subject from multiple directions, constructing 3D data and combining it with a background. The resulting scene can support viewpoints that are not identical to the position of a physical camera. That is different from switching between existing two-dimensional feeds.

The distinction is useful for understanding the promise: a viewer could inspect spacing between players from an angle suited to the moment. Our example is explanatory, not a claim that the demonstration offered a specific coaching or officiating tool.

Reconstruction also changes the questions an editor should ask. Does a moving hand retain a credible shape? What happens when one player obscures another? Does the ball remain continuous between frames? A compelling wide shot alone would not answer those questions. They belong in a production acceptance test.

Source notes: 2. Analysis and proposed examples are identified in the text.

Optical transport solves one part of the delay problem

NTT’s technical description explains how APN uses optical paths to reduce processing associated with repeated electrical-to-optical conversion. It presents low delay and low fluctuation as important properties for remote applications. That reference explains the network concept; its earlier demonstrations are not measurements of Canon’s September event.

Our engineering interpretation is to keep transport delay separate from end-to-end delay. A spectator sees the result only after capture, reconstruction, transmission and display have all contributed. Improving the connection cannot, by itself, establish the total time between a movement on court and its appearance in the headset.

A useful supplier report would therefore state the measurement endpoints. “Network latency” and “camera-to-display latency” should not appear as interchangeable labels. It should also show variation over the event, because an occasional long stall can matter even when a typical value looks good.

Source notes: 3. Analysis and proposed examples are identified in the text.

Moving equipment is not the same as removing it

Lumacta sees a potentially useful production model here: keep capture at the event and place reconstruction resources where they can be operated and maintained effectively. However, that is a hypothesis about operational value, not a saving demonstrated by the published announcement.

For a venue operator, we would compare the complete cost of cameras, installation, calibration, connectivity, computing, staffing and viewing equipment. Moving one stage off-site may change venue requirements while adding dependency on a particular connection. A fair comparison keeps both sides of that trade-off visible.

The commercial question is also one of utilisation. Equipment serving several events could justify a different investment from hardware used occasionally. We have no pricing, booking or capacity data for this system, so assigning a payback period would be invented precision.

Source notes: 1, 3. Analysis and proposed examples are identified in the text.

What a broadcaster should ask to see next

Our proposed test would use an uninterrupted passage of play, not only selected highlights. Compare the reconstructed action with the original camera views, including fast movement, occlusion and changes in lighting. Record the reconstruction and viewing settings so a second test is comparable.

Next, examine behaviour when a camera or network path is unavailable. Does the picture degrade, freeze or switch to a fallback? Does the production team receive a clear indication? These are questions for a controlled, authorised evaluation; they are not failures we observed in Canon’s demonstration.

Finally, ask viewers to complete a specific task, such as following a player away from the ball. Free viewpoint is an interface capability, not automatically a better way to understand a game. A conventional director-selected feed may still be preferable for some audiences and situations.

Source notes: 2, 3. Analysis and proposed examples are identified in the text.

The next milestone is a measured production service

From an imaging-systems perspective, the interesting advance is the separation of where a scene is captured, computed and experienced. It creates an architectural possibility. It does not establish that every venue can reproduce the experience economically or with the same quality.

The strongest follow-up would combine sustained image quality, end-to-end timing, failure recovery and cost per event in one documented trial. Reporting only the most flattering measure would make the result harder to evaluate, even if that measure were accurate.

For creators, this is a workflow worth watching rather than a product to order on the strength of a demo. The useful question is no longer just whether live action can become an explorable 3D scene. It is whether that scene can be delivered predictably enough for the production that needs it.

Source notes: 1, 2, 3. Analysis and proposed examples are identified in the text.