NovSU Master’s Student Develops a “Pocket Referee” for Armwrestling Competitions
NovSU Polytechnic Institute master’s student Tikhon Yakovlev is developing a mobile app that automatically detects rule violations in armwrestling.
Powered by artificial intelligence, the system analyzes athletes’ starting positions and alerts the referee to false starts or improper hand placement. The project is being developed through the Student Startup program run by the Foundation for Assistance to Small Innovative Enterprises and has reached the working-prototype stage.
Real-Time Analysis
The algorithm uses two modern computer vision architectures, MobileNet and YOLOv8-Pose. These neural networks specialize in object detection and tracking the human skeleton directly in video streams. The app tracks wrestlers’ shoulder, elbow, and wrist positions in real time, checking their geometry against the official rules of the World Armwrestling Federation.
The project’s key advantage is that all processing happens locally. The neural network analyzes the video stream directly on the mobile device, so the app doesn’t need a constant internet connection or access to cloud services. That means it can work even when there’s no network coverage.
A Smartphone Instead of a Camera System
Traditional video assistance systems for referees, such as soccer’s VAR (Video Assistant Referee), require dozens of synchronized high-resolution cameras and powerful data-processing servers. These setups cost millions of rubles, putting them out of reach for regional federations and organizers of local tournaments. The new development aims to make the technology far more accessible.
The “pocket VAR” needs only a standard modern smartphone mounted on a tripod beside the armwrestling table. This digital referee fits in an official’s pocket and is ready to use in seconds.
The system doesn’t replace the human referee. Instead, it acts as an intelligent assistant, flagging potential violations and highlighting the relevant areas on the device’s screen. This gives referees objective information to help them make sound decisions in disputed situations. The decision to disqualify an athlete or issue a warning, however, always remains with the match referee.
The World’s First Armwrestling Dataset
Training neural networks for highly specialized tasks often runs into the same obstacle: a lack of annotated data. No ready-made, publicly available image libraries for armwrestling existed, so the developer had to build what he describes as the first specialized dataset of its kind.
The dataset contains thousands of frames from real training sessions and competitions. Yakovlev manually annotated the athletes’ skeletal keypoints and linked them to geometric reference points on the table. Training also accounted for situations that conventional pose-recognition algorithms struggle with. The neural network even learned to determine hand positions correctly when the athletes’ hands overlap during a tight grip.
Using the table’s starting-position markers allows the software to understand the hands’ spatial orientation relative to the permitted boundaries. That matters when detecting elbow fouls and elbows lifting off the designated pads before the match begins. The algorithm also filters out visual noise effectively, ignoring brightly colored sportswear, long sleeves, and variations in indoor lighting.
From Local Tournaments to International Competitions
The project is currently undergoing pre-pilot testing. The next step will be field trials at official tournaments, with methodological support from the Armwrestling Federation of the Novgorod Region. Integrating the app into actual refereeing procedures is still ahead, and will require establishing clear protocols for how the digital assistant and human officials work together.
The project fits into a broader global trend toward digitizing sports. In 2024, the International Olympic Committee adopted its Olympic AI Agenda, which encourages the use of neural networks in competition. FIFA is actively testing accessible Football Video Support systems for grassroots tournaments. Meanwhile, researchers in Novgorod are developing an AI coach for skateboarding, helping the university build a broader cluster of sports technology projects.
The technology could eventually be adapted for other sports, too. Algorithms that track athletes’ starting positions could be useful in powerlifting, weightlifting, and various combat sports. A successful rollout at the regional level could pave the way for commercialization through licenses sold to sports clubs. Over time, the affordable “pocket VAR” could also attract national federations in other countries looking to make officiating more objective despite limited resources.
