AiDi Delivers Real-Time Record Line Overlay at the Tokyo Marathon

AiDi # image recognition# Object Detection

2026.03.12

Author

Yuki Miura

On March 2, 2026, the Tokyo Marathon 2026 was held with a record-high field of 39,000 runners. Beyond expectations for new world and national records, the race carried added significance as a qualifier event for the Asian Games scheduled for this autumn, as well as for the Marathon Grand Championship (MGC) on the road to the 2028 Los Angeles Olympics, placing an exceptional spotlight on records.

To convey the intensity of the race as it unfolded in real time, Nippon TV utilized the image recognition feature of AiDi, its proprietary AI technology. World-record and national-record pace lines were overlaid directly onto the course, enabling instant visual comparison with the leading runners and providing viewers with a clear and intuitive understanding of race progression.

Real-Time Processing of Three Locations with a Single PC, Reducing Equipment and Costs

Conventional real-time compositing has required detailed numerical data synchronized with camera lens movements, such as zoom levels and focus positions. To obtain this data, specialized equipment needed to be installed and operated at each broadcast location. As a result, attempting compositing at multiple points along the course significantly increased both equipment costs and staffing requirements.

In contrast, AiDi’s image recognition technology analyzes the video itself to perform CG compositing, eliminating the need to physically connect camera lenses and equipment to extract lens data. This technical simplification makes it possible to remotely process video from three locations simultaneously using a single PC, achieving substantial reductions in both equipment and personnel.

AI-Driven Automatic Masking for
Real-Time Runner and CG Depth Adjustment

When CG graphics are conventionally displayed on the course, they tend to overlap with runners or vehicles in the foreground, reducing visual clarity.

To address this issue, image recognition is used to detect people and vehicles in the foreground in real time, and a masking process is applied to recreate natural depth, allowing record pace lines to pass behind runners and lead bikes. This eliminates visual inconsistencies and enables clear, immersive visuals without compromising the realism of the race.