WeisonWEileen/yolo_mediapipe

This is a project that combine yolo-v8 and mediapipe to assist the daily life of patients with hemiplegia grasping

YOLOv8 + Mediapipe to assist the daily life of patients with hemiplegia grasping

work flow chart

test on patient

test demo

3D online hand pose visulization

how to run: open a terminal window

python3 mediapipe_3d.py

open another terminal

python3 extract_mediapipe_hand.py

FSM control algorithm

how to run the container

Firstly, we need to build the image using the Dockerfile(you should run the following command in the parent folder of Dockerfile )

docker build -t yolo-medpip-eletron

Then run the container

sudo docker run -dit \
--name=yolo_mediapip_eletronic \
--privileged  \
-v /dev:/dev \
-v /tmp/.X11-unix:/tmp/.X11-unix  \
-e DISPLAY=unix$DISPLAY \
-w /usr/src \
--net=host \
--ipc=host \
--gpus all \
yolo-medpip-eletron

This is the plotting call stack

flowchart TB;

    subgraph distance computing
        id6
        id10
    end

    subgraph yolov8
    direction TB
        id1[result]--select bottle-->id2[annotator]-->id3[boxes]--pred_boxes_show_boxes-->id4["box_label() "]-->id5["cv2_rectangle() cv2_circle()"]
        id3-->id6["box_center"]
    end

    subgraph mediapipe
    direction TB;
        id8[result]-->id9[hand_landmarks]-->id10[index_9 landmark]
        id9-->id14["cv2.circle()"]
    end
    

Loading

Tips:

• add name property name_handler in the Box class for the result extract in the main function, which is assigned in the Results.plot().

  d.name_handler = label
  

• and the final 2D box center is also assigned in the Results.plot().annotate.box_label() function.

  self.center_handler = 0
  

@ TODO

• 读对面的，记录电刺激参数的帧末尾处理，记录，下次读进来作为，读杯子的宽度

Appendix

here is the official hand detection output

combine.py 是最终的运行文件