👍 HGR - Hand Gesture Recognition

HGR is a hand gesture recognition system built with Google MediaPipe.

🚧 Under Development

⚙️ Usage

git clone https://github.com/surajkareppagol/HGR
cd HGR

python3 -m venv venv

source venv/bin/activate

pip install -r requirements.txt

python3 hgr/main.py

deactivate the virtual environment.

deactivate

✅ Requirements

• 🐍 Python3.7+
• ✨ mediapipe
• ✨ mediapipe-model-maker
• ⚡ rich
• 🌟 ttkbootstrap
• 🖱️ pyautogui

Mediapipe requires Python versions between 3.7 and 3.10. In Windows there are some problems with mediapipe installation, so it is recommended to install these using pip.

Recent changes in tensorflow broke the model maker, so use the following to install mediapipe-model-maker.

pip install "keras<3.0.0" mediapipe-model-maker

↔️ API Class

The hgr/mp/api.py provides a API Class, that has following methods.

• get_gestures()

  Returns a list of available gestures.

• get_landmarks_image()

  Returns landmarks from a image passed as argument.

• draw_landmarks_image()

  Draws points on landmarks and connects them.

• get_landmark_image()

  Returns landmarks for a specific point, between 0 - 20.

  

• get_gesture_image()

  Classifies the gesture from image, and returns it.

🖥️ GUI

🎯 Model Maker Scores

╔═══════════╤══════════╤════════════════════╗
║ Iteration │ Gesture  │       Score        ║
╟───────────┼──────────┼────────────────────╢
║     1     │   none   │ 0.8584573268890381 ║
║     2     │   none   │ 0.568006694316864  ║
║     3     │  paper   │ 0.9921259880065918 ║
║     4     │  paper   │ 0.991919994354248  ║
║     5     │  paper   │ 0.9897699356079102 ║
║     6     │  paper   │ 0.9981831908226013 ║
║     7     │  paper   │ 0.9804528951644897 ║
║     8     │   none   │ 0.893876314163208  ║
║     9     │  paper   │ 0.9854918122291565 ║
║    10     │   none   │ 0.8718634247779846 ║
║    11     │   none   │ 0.760579526424408  ║
║    12     │ scissors │ 0.5368748307228088 ║
║    13     │   none   │ 0.8792762160301208 ║
║    14     │   none   │ 0.7174986600875854 ║
║    15     │   none   │ 0.9842039942741394 ║
║    16     │   none   │ 0.9359919428825378 ║
║    17     │   none   │ 0.948810338973999  ║
║    18     │   none   │ 0.9333010911941528 ║
║    19     │  paper   │ 0.6957719326019287 ║
║    20     │ scissors │ 0.9800419211387634 ║
║    21     │ scissors │ 0.6627081036567688 ║
║    22     │ scissors │ 0.9795249104499817 ║
║    23     │ scissors │ 0.9814074039459229 ║
║    24     │ scissors │ 0.9789888858795166 ║
║    25     │   none   │ 0.9442585706710815 ║
║    26     │ scissors │ 0.959807276725769  ║
║    27     │   none   │ 0.9233239889144897 ║
║    28     │   none   │ 0.9254647493362427 ║
║    29     │  paper   │ 0.8687804937362671 ║
║    30     │  paper   │ 0.8441806435585022 ║
║    31     │          │ 0.8685176372528076 ║
║    32     │   none   │ 0.965121328830719  ║
║    33     │   none   │ 0.9680899977684021 ║
║    34     │ scissors │ 0.9052069187164307 ║
║    35     │ scissors │ 0.8518658876419067 ║
║    36     │ scissors │ 0.9516168236732483 ║
║    37     │ scissors │ 0.9168112874031067 ║
║    38     │ scissors │ 0.9222723245620728 ║
║    39     │ scissors │ 0.9808273315429688 ║
║    40     │ scissors │ 0.9677468538284302 ║
║    41     │   none   │ 0.8759973049163818 ║
║    42     │   none   │ 0.8951284885406494 ║
║    43     │   none   │ 0.5676480531692505 ║
║    44     │   none   │ 0.8585500717163086 ║
║    45     │   none   │ 0.9657561182975769 ║
║    46     │   none   │ 0.8343509435653687 ║
║    47     │   none   │ 0.8246175646781921 ║
║    48     │   none   │ 0.9299445152282715 ║
║    49     │   none   │ 0.9493988156318665 ║
║    50     │   rock   │ 0.6069502830505371 ║
║    51     │   none   │ 0.9981077909469604 ║
║    52     │   rock   │ 0.526486337184906  ║
║    53     │   none   │ 0.8241088390350342 ║
║    54     │   none   │ 0.547508180141449  ║
║    55     │   none   │ 0.7531124949455261 ║
║    56     │   none   │ 0.9233627915382385 ║
║    57     │   none   │ 0.9331709742546082 ║
╚═══════════╧══════════╧════════════════════╝