A Copenhagen startup developed a deep learning-based camera system that can detect where the action is on the soccer field, and automatically zoom and follow the ball – just like how a camera operator would do.
“Today, less than 1 percent of all football matches are recorded,” says Veo co-founder and CEO Henrik Teisbæk. “This is because in order to record a football match properly, you need a cameraman to be filming from an elevated position for 90 minutes, and then be able to edit the footage afterwards. Most teams simply don’t have the resources required for this, meaning that millions of goals and unique footballing moments are never viewed or shared by the players”.
Using CUDA, TITAN X GPUs and the cuDNN-accelerated TensorFlow deep learning framework, the small startup trained their deep neural networks on over one million soccer-related images to automatically track the ball and players with consistency and precision.
“We have spent nearly two years developing the technology to detect where on the pitch a camera operator would point and zoom,” Teisbæk explains.
Combined with Veo’s camera mount consisting of two 4K cameras that produce a full 180-degree panoramic view of the entire filed, their trained AI technology takes advantage of compute resources in the cloud to follow the action.