Final Project of Applied Deep Learning 2017 in NTU.
Track the movements of NBA players and map them onto a tactic board.
Extract frames from NBA highlight video taking advantage of ffmpeg.
p.s. In warrior_vs_jazz/
directory
Use YOLO or Faster-rcnn to detect players in each frame.
p.s. In yolo/
directory
Tandform each bounding box into histogram vector and label three bounding box to classify players' corresponding teams.
p.s. In player_classify/
directory
Utilize court line to map frame to tatic board.
p.s. In mapping/
directory
For each frame, use their former or latter frame to delete rebundant point and do track smoothing.
p.s. In index2court/
directory
Convert frames into video with ffmpeg.
p.s. warroirs_vs_jazz.mp4
sh train.sh
cd index2court/
python3.6 player_track_warriors.py
p.s. Output frames will store in index2court/game_out2/
.
Step1 : Object detection (YOLO)
Step3 : Mapping
1, Line detection and DBSCAN:
- cython
- protobuf
- cv2
- ffmpeg
- matplotlib
- numpy
- pandas
- pickle
- python3.6
- sklearn
Team : ADL 躺分仔
- 李承軒 B03902009 https://github.com/Spicy30
- 陳雋 B03902033 https://github.com/falloutboyrocks
- 顏廷宇 B03902052 https://github.com/y95847frank
- 紀典佑 B03902059 https://github.com/dianyo
- 邵楚荏 B03902090 https://github.com/nickshao
[1] S. Ren, K. He, R. Girshick, and J. Sun. Faster r-cnn: To- wards real-time object detection with region proposal net- works.
[2] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You Only Look Once: Unified, Real-Time Object Detection.
[3] Evan Cheshire, Cibele Halasz, and Jose Krause Perin. Player Tracking and Analysis of Basketball Plays.