1 of 17

VolleyVision:�Spatio-Temporal Analysis of Volleyball Data for

Complete Tracking

Author: Shakhansho Sabzaliev

Supervisor: Artyom Kulakov

2 of 17

Problem Statement

🏐

🏐

3 of 17

Lack of ML & CV applications to the field of volleyball

​

4 of 17

5 of 17

Enrich my knowledge and develop understanding of ML algorithms

6 of 17

Contribute to open-source community

7 of 17

Solution

8 of 17

VolleyVision

  • Web App
  • GitHub repository
  • Object Detection Datasets

    • Volleyball & Player Detection
    • Court Tracking
    • Action recognition
    • Statistics / Insights
    • Highlights

9 of 17

Methodology

10 of 17

I STAGE�VOLLEYBALL TRACKING

II STAGE�PLAYERS TRACKING AND ACTION RECOGNITION

III STAGE �COURT TRACKING

IV DELIVERY

  1. Research existing algorithms
  2. Data Collection & Annotation
  3. Training & Testing
  4. Evaluation

11 of 17

I STAGE - VOLLEYBALL TRACKING

Models:

  • RoboFlow & DaSiamRPN
  • yoloV7-tiny

Dataset:

  • Volleyball Detection Dataset (25k)

12 of 17

II STAGE - PLAYER DETECTION & ACTION RECOGNITION

Models:

  • Yolov8M
  • RoboFlow

13 of 17

Players Detection

Action Recognition Dataset

Datasets

14 of 17

III STAGE - COURT DETECTION

Models:

  • RoboFlow

Dataset:

15 of 17

Demo & Results & Limitations

GitHub Repository

Volleyball Tracking

Player Tracking

Court Tracking

Datasets

Current Models

Web App (local)

16 of 17

`

17 of 17

Questions?