MASK.AI logo

Tracking cars is hard and tracking cars at 300 km/h is even harder. And by tracking we don't mean bounding boxes, we mean full 6DOF poses and internal and external camera parameters, all in real-time and with high reliability. This is exactly what MOTO.AI is all about.

Here is a little sample. Imagine your favorite race here, MOTO.AI works with any car racing event. What you see are raw 3D data points streamed from our engine. Given these data points, it is possible to render arbitrary graphics and blend it with the input video. It is also possible to view the scene from completely different angle and inspect racing incidents. Or take it as raw measurements and fuse it with physical sensors to further improve your on-track positioning system. Possibilities are endless.

The potential of MOTO.AI was recognized by EPIC MegaGrant. And we are thrilled to push our project further because of them. Current version of MOTO.AI is packed with features, and more are being developed every day.


Vision-based, real-time, on-track 3D positioning data with game-changing performance.

Takes advantage of external sensors if available (e.g. car telemetry, inductive loops, GPS)

Car detection from 20x20 pixels at 60 fps in HD1080 video stream.

Car pose estimation with 1 degree standard deviation.

Car translation estimation with 10 cm standard deviation.

Rigged 3D car models for all major racing series.

Modeling of car wheel spin, turn angle, rear-axis offset.

Real-time rendering of training data, realistic augmentation of occlusions.

Data augmentation with motor-port specific noise factors.

Highly sophisticated 3D annotation tool, with embedded VFX solvers.

Physics-driven constraints on the resulting car trajectories.

Extensive training set of pixel-accurate aligned 3D models.

Custom neural network architecture optimized for 3D object detection and pose estimation.

Driver recognition using timing information.

Driver recognition using visual recognition.

Streaming of estimated data points into external rendering servers (VizRT).

Tracking through heavy occlusion and complete disappearance.

Resistance to camera jitter, rain and fog.

Modeling of camera distortion parameters.

End-to-end training of the whole pipeline

Procession of live video input through SDI interface


Watching a race is quite off-putting for people who don't know much about the sport. After all, it is just cars going in rounds. Well not quite that. Each driver has its own story, and knowing the background turns the sport into much more. Driver tagging is a technology, that allows to associate cars with driver faces and make the experience much more personal. The race will become a story, not just fast images of cars.


There is tons of data points flying in and out during the race. Bringing all of them to the audience in an easy-to-digest way is quite a challenge. MOTO.AI has developed a technology that allows to balance the information content and clutter on the screen. The graphics is directly driven by the motion and 3D location of elements on the screen, so it is always clear what the message is. Graphics can be thus flipped faster and maintain clean look expected from AI driven interface.