Futsal-Image-Playing-Analysis-Service

work added : webRTC(by @chaeyeon4u), kinesis(by @hkjs96) , MOTR(by @heosuab)

You can watch demo video from here : https://youtu.be/bA6G8tl7K14

Goal

Stitching multi-videos into an entire view of futsal game and analyze it by tracking players and a ball

촬영앱 유저 인터페이스

로그인

회원가입

메뉴

Creating Top down view Video

We expect that we can take the top-down view by this 4m tripod at all type of futsal field

WebRTC(Real-Time-Communication) for monitoring and Video Record

---

• Web RTC(Real-Time-Communication) with Kotlin
• P2P Network
• Create Key
• Video Record
• Login / Signup

실시간 모니터링

< Create Room Key >

사용자가 입력한 키가 유일한 키이면 방 생성.

키가 다른 사람이 만든 키와 중복될 경우 방 생성 불가.

< WebRTC Start & Record Start >

< 카메라 전환과 RTC & Record >

영상 녹화 결과 영상 (일부)

AWS Kinesis Streaming

Monitoring scenes from two device on a single page

Stitching and Tracking two videos by Tkinter GUI Program (step1~step )

Step1: Open Left, Right Source Frame

Step2 : Check a Pre-Stitched Frame and cut a video

We need two decide 6 points. 2 points for cropping and 4 points for converting coordinates to the real data.

Step3 : Stitching

Version 1.0 : Perspective transform

Version 2.0 : User defined homography

Version 3.0 : Warping by Cylindrical Laplcian Blending

Step4: Tracking

We use DeepSORT algorithm to track multi-objects(players,ball) and produce csv coordinate file

Version 1.0 : DeepSORT + YOLOv3

Version 2.0 : DeepSORT + YOLOv4

Version 2.1 : DeepSORT + YOLO4

Version 3.0 : MOTR(Improved ID switching issue)

Data Visualization

We visualzied and calculate heatmap, minimap(video), average position map, activity, spped. We described those on the Flask webServer