Sports Broadcast Processing & AI Analytics

Our client is an international company specializing in the production and distribution of sports media content. The company’s portfolio includes live broadcasts and clips of various sports: football, basketball, hockey, tennis, and MMA. The main goal is to increase audience engagement through intelligent editing of broadcasts and accurate accounting of visual content, including graphics, advertising, and game events.

The company approached us with the aim of improving its workflow pipeline: an intelligent system for analyzing and modifying video streams in real-time, without the need for manual editing and post-processing of each match, was required.

Problem / Task

Key challenges faced by the client:

• Lack of automatic comparison of broadcasts to identify graphics overlays.
• Inability to accurately recognize and classify advertising elements and visual layers.
• Inefficient content adaptation for mobile platforms and social networks.
• Delays in creating highlights and reactive content.
• Lack of an automatic system for detecting key game events.

Solution

We developed a universal CV system and video analytics modules, which included:

• Automatic comparison of two broadcasts: highlighting differences, graphic layers, advertising.
• Detection of overlaid graphics with the identification of its type (score, timer, logo, advertising).
• Semantic segmentation of the field and advertising zones (including virtual inserts).
• Event detection (goal, shot, foul, dangerous moment) with time and coordinate binding.
• Auto-cropping and changing the video format for vertical social media feeds (from 16:9 to 9:16).
• On-the-fly generation of modified content depending on events and marketing tasks.

All modules were implemented as isolated microservices with REST and WebSocket APIs, compatible with the client’s production environment.

Process

The full implementation cycle took 7 months and included:

• Analysis of graphics and video formats from different sources.
• Building a dataset for training and validation.
• Development and testing of CV models and event heuristics.
• Integration into the client’s broadcast platform.
• A/B testing in live matches.
• Optimization of latency and GPU load for live scenarios.

The team used a flexible CI/CD process and a production quality monitoring system.

Result

• Up to 95% accuracy in detecting overlaid graphics.
• 87% accuracy of event detection for key moments (goal, shot, foul).
• Auto-cropping and video repackaging with a delay of less than 2000 ms.
• Over 70% reduction in content preparation time for social networks.
• A 500% increase in the number of videos in 9:16 format without additional staff.
• Increased viewer engagement — 60% more likes and reposts compared to manually created clips.

Features / Difficulties

• Working with various graphics formats — from national league broadcasts to online tournaments.
• Automatic determination of the field area and advertising zones at different shooting angles.
• Combating false positives in event identification — hybrid algorithms (visual + audio track analysis) were developed.
• Integration without stopping production: the system was deployed in parallel in “observer” mode.
• Automatic format correction for aspect ratio violations in the original video.

Interesting Facts

• More than 80 matches with commentary and graphics from five countries were used in training.
• The system independently recognized the overlay of a sponsor’s logo not declared on air.
• Event detection was tested even on low-quality streams with minimal delay.
• The auto-crop module is adapted to TikTok, Instagram Reels, and YouTube Shorts — with optimization of the focus area along the ball’s trajectory.

Feedback

“We got a tool that completely transformed our content creation process. Now everything works faster, more accurately, and most importantly — without the participation of editors. The team quickly understood our tasks and offered solutions that really work on air.”

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.