Tennis Registration v3

Prepare the environment

1. DeepSORT environment

cd YOLOv9_DeepSORT
pip install -r requirements.txt

2. YOLOv9 environment

cd yolov9
pip install -r requirements.txt

3. UI environment (cpp-OpenCV & sysv_ipc)

cpp-OpenCV v4.7.0 in Ubuntu (Here is the tutorial)

pip install sysv_ipc

Demo

You can use the code below to simply run the demo.

python tennis_registration.py

After running all frames, you will see the result in the terminal. This is the position of every player in the image coordinate.

[[      31.76      8.2214]
 [     22.868      10.646]
 [     11.693      7.4278]
 [     6.3028      9.9999]]

Or if you want to see the result, you can run the ui at the same time.

cd ui
./shr_tennis_ui

Then you will see like the image below.

Usage

You can use your own params to run the program with the code below.

python tennis_registration.py --video <str> --conf <float 0~1> --blur_id <int> --class_id <int> --cam_id <int>

Params

If you just simply test its performance in different perspectives, just change its video and cam_id params.

Supplements

① Stream Mode

If you need to use Stream Mode, pls apply this two code in tennis_registration.py after you connect the camera. Then input ‘0’ to video param. And don’t forget to change a correct camera index to calculate homography matrix. (You can rewrite cam_dict in object_tracking_stream to fix the correspondence between cameras and indexs)

# from YOLOv9_DeepSORT.yolov9.object_tracking_stream import person_track_stream

# person_track_stream(FLAGS)

Because I cannot connect my computer with cameras in the stadium, I commented them out while debugging the code. (If there are bugs when using Stream Mode, pls contact me…)

② Differences between object_tracking_stream and object_tracking

The code of video stream is in object_tracking_stream.py, and the code of video file is in object_tracking.py They use different methods to get frames: class PYLON_SHM and VideoCapture function in OpenCV.

③ Video samples and cam_id

The indexs of videos in the data folder are corresponding to the indexs of cameras.

video: video_0.mp4 -> cam_id: 0
video: video_1.mp4 -> cam_id: 1
video: video_2.mp4 -> cam_id: 2
video: video_3.mp4 -> cam_id: 3

References

YOLOv9_DeepSORT

Learn Homography: 相机标定系列（二）单应矩阵

Homography Applcation: OpenCV-Python单应矩阵（Homography Matrix）应用——更换广告牌