Gal Oz and Miky Tamir had already worked together on sports tracking at SportVu. When they started Pixellot in 2013, Oz saw a change in the raw material of their work: camera resolution was improving while the cost of capturing those pixels was falling. That made it possible to start with a wide view and decide where to look from the footage itself.
“let’s capture the full game all the time, let’s automate production”
The first version still needed a person
Oz told TVBEurope that the first development effort took about a year. Some early workflows still needed someone in the production process. Keeping the whole playing area in frame solved one problem. Choosing which part viewers should see at any moment was a different job, and it became the part Oz and Tamir focused on automating.
Learning where to look
A camera following basketball has to make different choices from one following soccer. Oz described testing new sports in beta, collecting more footage and refining the system. For him, the wide shot was the starting point. The work was learning enough about each game to let a viewer follow what mattered as play moved.
AI in the product · November 13, 2026
Oz’s early work was about making it possible to cover a game with fixed cameras and software that could choose where to look. Pixellot says it trains its tracking separately for each sport. That allows a basketball broadcast and a soccer broadcast to follow different patterns of play.
The automation eventually changed how Pixellot could sell the service. In describing its five-year restructuring with PlayOn, Pixellot said its revenue-sharing model could remove the upfront cost of the camera. The engineering decision to automate coverage opened up a different business arrangement for venues and partners that might otherwise have had to buy the hardware first.