Computer Vision Subsystems for Racket Sports products (Tennis, Padel)

Our client is an international company operating in the sports analytics and broadcasting segment, specializing in racket sports, including tennis and padel. The company provides a technological platform for video broadcasts, in-depth match statistics, player analysis, and automated content production.

The client was already using cameras to record matches but experienced difficulties with accurate ball and court line detection, as well as automatic recognition of game events. A comprehensive computer vision system was needed that could operate in real-time and adapt to the dynamic conditions of padel courts.

Racket Sports Computer Vision Demo

See selected examples of ParadigmaST computer vision subsystems for racket sports, including Tennis, Padel, and Badminton.

The video combines demonstration samples from multiple racket sports implementations and technical experiments. It is intended to showcase our broader computer vision capabilities and does not represent a single client project or a single production system.

The examples demonstrate different combinations of court detection, player tracking, ball tracking, pose estimation, event recognition, and video analytics that can be adapted to specific racket sports products and camera setups.

Problem / Task

The main technical challenge was not individual object detection, but maintaining stable tracking and event recognition across real match conditions. Fast ball movement, partial occlusions, reflections from glass, changing lighting, and players temporarily leaving the camera view required several computer vision components to work together as one system.

The main tasks set by the client were:

• Accurate detection of markings and lines on the court with partial overlaps and glares.
• Tracking the ball and players at high speed, including during sharp changes of direction.
• Recognition of game events: serve, out, bounce, change of sides.
• Tracking and identification of players even when they go out of sight.
• Support for padel-specific cases, including a player changing clothes during a match.
• Real-time operation and compatibility with the client’s existing video stream.

Solution

We developed and implemented a modular CV system, including:

• Calibration and semantic detection of the court and markings (including dynamic detection when changing cameras).
• High-precision ball tracking with 2D/3D trajectory reconstruction.
• A separate module for bounce detection, taking into account the surface (glass, net, carpet).
• Recognition of the type of hit based on pose estimation + trajectory.
• Segmentation and pose estimation of players on each frame.
• Automatic identification of players, even after going out of the field and changing clothes.
• Event detection: out, serve, rally, end of point.
• Microservice architecture with REST/WebSocket API for integration with the client’s platform.

Instead of building one monolithic model, the solution was designed as a modular computer vision pipeline. Individual components could be validated, improved, and integrated independently while sharing calibration, tracking, and event data through a common processing layer.

Racket Sports CV Capabilities

Beyond this specific client project, ParadigmaST develops reusable computer vision components and custom subsystems for racket sports products.

Depending on the sport, camera configuration, and product requirements, these may include:

• court and key-point detection;
• camera calibration;
• player detection and tracking;
• ball detection and trajectory tracking;
• pose estimation and movement analysis;
• serve, rally, bounce, hit, and other event recognition;
• player identification and re-identification;
• real-time and post-processing video analytics;
• structured data output for integration with sports platforms, dashboards, and applications.

Process

The project lasted 7 months, and the stages included:

• Researching the specifics of padel games and preparing custom datasets.
• Training models for various lighting conditions and camera configurations.
• Validating the accuracy of tracking and event recognition in manual tests.
• Integrating modules with the client’s production stack via CI/CD.
• Supporting deployment on edge devices with low latency.

Result

The system was validated on recorded and live match footage across different court and camera conditions. The final performance depended on video quality, camera position, lighting, and the specific analytics module, but the production configuration achieved the following results.

• Up to 95% tracking accuracy for ball and player tracking in validated court and camera configurations.
• Over 92% accuracy for bounce and out-event detection in tested scenarios.
• Event detection latency below 200 ms in the target deployment configuration.
• Automatic player re-identification after temporary exits from the camera view and selected appearance changes.
• Match-report preparation time reduced from approximately 3 hours to around 10 minutes within the client’s workflow.
• The system was integrated into the client’s existing platform without interrupting active video streams or platform operations.

Features / Challenges

The project required:

• Solutions to non-standard padel cases: players leaving the field and returning.
• Identification when a player changes their appearance (e.g., taking off a shirt).
• Working with reflections from glass walls, unstable lighting, and glares.
• Models adapted to the conditions of both outdoor and indoor courts.
• All analytics work on the client’s side without transmitting video to the cloud.

Interesting Facts

• To train the models, we manually marked up more than 100 padel matches.
• The system successfully recognizes moments when a player takes off their shirt and returns to the game.
• Using pose estimation helped predict not only the type of hit but also the probability of winning the rally.
• The modules are used to automatically create highlights and match clips.

Mykhailo Smorodin

CEO at Paradigma.ST, Computer Vision and AI in Sports Tech

Motivated and entrepreneurial software engineer with a passion for artificial intelligence, computer vision, and sports. Skilled in developing innovative solutions and driving projects from concept to completion. Eager to leverage my technical expertise and business acumen to create impactful advancements in technology.