U-GO Logo

U-GO

AI-Powered Soccer Coaching Assistant for Nigerian Youth

Role

Computer Vision Engineer · UX Researcher · AI/ML Developer

Stack

YOLOv8 · MediaPipe · OpenCV · Flask · Roboflow · Python

Team

Fatima Mamu, Prachi Mehta, Valentina Arango, Yiko Li

Project Overview

The Problem: Nigeria's Coaching Disparity

Soccer offers young Nigerians a pathway for socioeconomic advancement, but a critical infrastructure gap exists. While talent is universal across Nigeria, access to quality coaching is not, leaving potential stars without the guidance needed to develop their skills.

The Reality: Most schools assign general PE teachers as soccer coaches due to financial constraints, creating a systematic barrier to proper tactical and technical development for 98% of young players.

The Solution

U-GO is an AI-powered coaching assistant designed to democratize access to quality soccer training in Nigeria, where only 2% of young players have access to specialized coaching. The platform allows coaches to upload player videos for automated skill analysis and receive personalized training recommendations.

User Research & Technical Requirements

Field Research with Nigerian Soccer Coaches

Research Map of Nigeria

Research conducted across Lagos, Kaduna, and Abuja - 3 cities, 12 interviews, 5 observations

Research conducted across Lagos, Kaduna, and Abuja to understand current coaching workflows, technology adoption patterns, and critical pain points in youth soccer development.

Key Research Insights

Technology Adoption

67% of coaches used smartphones for basic tasks but were hesitant about complex apps due to limited technical training

Feedback Preferences

Coaches strongly preferred visual performance indicators and simple charts over numerical scores and detailed statistics

Time Constraints

Average training sessions lasted 90 minutes with only 10-15 minutes available for individual player analysis

Connectivity Issues

Poor internet reliability required offline-capable solutions, with 40% of training locations having intermittent connectivity

Stakeholder Validation

Direct quotes from our research revealed the depth of the challenge:

"They love to play soccer, but really the school has not provided a situation where there is formal training for these children, especially the enthusiastic girls. There is just a common PE coach for all."

— Aminatu Zakari, School Principal

"No system to create value for soccer enthusiasts at the grassroots level. It's a socio technical issue and looking at providing soccer mentorship to everyone can really impact a lot of kids in low income communities."

— Sadiq Katika, Soccer Player and Coach

My Technical Contributions

Computer Vision Pipeline Development

Built end-to-end analysis system:

Detection
Tracking
Tactical Mapping
Skill Quantification

Core Technologies:

YOLOv8

Custom trained models for player and ball detection across varied field conditions

MediaPipe Pose

Real-time player pose estimation and movement tracking

Homography Transformation

Field keypoint detection and perspective correction for spatial analysis

OpenCV

Video preprocessing, stabilization, and frame extraction

Technical Implementation

Player and Ball Detection Icon
  • Fine-tuned YOLOv8 models using Roboflow for robust detection across lighting conditions
  • Achieved 80% player detection accuracy and 75% ball detection accuracy in varied environments
  • Handled multi-player scenes with player isolation and tracking
Player and Ball Detection

Player and ball detection and tracking on one of the test matches we recorded in Bwari, Nigeria

Field Analysis Icon
  • Implemented automatic field line detection using edge detection and Hough transforms
  • Applied homography transformation for standardized tactical mapping
  • Generated heat maps showing player positioning and movement patterns
  • Converted pixel distances to real-world measurements using field dimensions
Tactical Mapping

Field key point detection + homography transformation on the harvard vs MIT match we recorded

Skill Quantification Icon
  • Ball Control Score: Duration and smoothness of ball possession
  • Spatial Awareness: Player positioning patterns using field mapping
  • Sprint Analysis: Distance coverage and speed calculations
Mobile App Interfaces

Usable feedback interfaces for coaches and actionable recommendations for players

Technical Challenges Solved

Multi-Player Scene Analysis

Challenge:

Isolating target player performance in crowded training scenarios

Solution:

Combined YOLOv8 detection with tracking algorithms and spatial consistency checking

Achievement:

High accuracy in correctly attributing actions to target players

Key Technical Learnings

Field Testing Results

Conducted live testing at Harvard vs MIT soccer game and conducted testing at Bwari, Nigeria, validating tactical mapping and player tracking systems under real match conditions

Computer Vision in Uncontrolled Environments

Real-world robustness requires extensive preprocessing and adaptive thresholding. Field sports analysis benefits significantly from geometric constraints and spatial context.

User-Centered Technical Development

Research-driven architecture leads to higher adoption. Offline-first design is critical for emerging markets. Visual feedback systems are more effective than numerical metrics for non-technical users.

Cross-Functional Collaboration

Technical specifications must align with UI constraints and AI system requirements. Iterative testing with end users is essential for practical computer vision applications.

U-GO project video