Multi-Sports Analytics
A two-pass sports-analysis pipeline combining computer vision, structured data and targeted analysis.
2025
Passion Infotech collaboration · IEEE INSECT 2026.
Two-pass, multi-sport analysis
Multi-Sports Analytics organizes video and structured player data into a skim-and-focus workflow. The first pass identifies the sport and maps relevant scene structure. The second selects events and frames for detailed sport-specific analysis. Football, basketball and tennis modules connect Python processing to Streamlit views.
Developed through academic industry collaboration with Passion Infotech, the work is associated with the IEEE INSECT 2026 publication.
Two passes, different responsibilities
- Video input
- Skim: sport & scene
- Focus: relevant events
- Structured analytics
- Dashboard
YOLOv8/OpenCV video processing connects sport-specific modules to JSON and Pandas-backed analysis.
Sport-specific analysis
- Football: dedicated analysis logic interprets the footage in field/game context instead of applying a generic video summary.
- Basketball: a separate module handles court-oriented events and player information within the same two-pass structure.
- Tennis: the analysis module uses the sport’s own event context and exports information for the dashboard.
- Shared outputs: structured JSON and Pandas data handling support summaries, player views and comparisons across analysis runs.
Engineering decisions
YOLOv8/OpenCV processing produces visual detections and frame information; Pandas manages tabular player and historical records. JSON outputs preserve extracted information for subsequent analysis and dashboard use. Broad recognition narrows the input for detailed processing instead of applying one undifferentiated pass to the entire recording.
The source-defined analytical path uses K-Means to group player feature profiles and PCA to summarize multidimensional performance features. These indicators remain tied to their inputs. Sport-specific modules preserve different event and court/field definitions, while video and historical data contribute complementary context. The pipeline exposes intermediate detections and structured statistics, connecting footage to features and then to readable comparative views.
Publication & collaboration
Sports Analytics with Optimized Agentic Two Pass Approach Using Multi-Modal Data
IEEE INSECT · 2026 · Faculty collaboration with Dr. Premanand P. Ghadekar ↗Professor & Head of Department of CSE (AI & ML), VIT Pune
Read the paper