TheLazy.Dev
Portfolio GitHub

Blog

python |machinelearning

Day3: Play Trex Game on Chrome Using Gestures

qviper

Viper

Posted on July 19, 2021

Day3: Play Trex Game on Chrome Using Gestures

Play Trex Game on Chrome By Gesture Using OpenCV and Mediapipe

Hello there surfer!
Since few days, I am thinking about some cool projects that can be done within some hours using Mediapipe and OpenCV in Python.
In this blog, I am writing about how can we play the popular trex game by only moving our fingers in front of the camera. Many of us have played this game but none of us were interested to play. 🤣🤦‍♂️🤦‍♂️ Well in this blog, we are going to play with with full intention.

This blog is the part of the series #7DaysOfComputerVisionProjects. Links to the blogs and videos of each projects are:

  1. Real-time Background Changing: Video | Blog
  2. Air Mouse: Control Mouse with Gestures Video | Blog
  3. Play Trex Game With Gesture Video | Blog
  4. Auto Dino: Play Trex Game Automatically Video | Blog
  5. Gesture Based Writing Video | Blog
  6. Game: Kill The Fly Video | Blog
  7. Gesture Based Calculator Video | Blog

Prerequisties

  • Mediapipe: Install it using pip install mediapipe.
  • OpenCV: It will be installed by default while installing mediapipe.
  • Keyboard: Not the physical one because we are simulating key events using gestures. pip install keyboard.

Once installed make sure you can use them. Just import them and see if any error pops up.

import mediapipe as mp
import cv2
import numpy as np
import keyboard
Enter fullscreen mode Exit fullscreen mode

Using keyboard package is pretty easy just like mouse package. For test, we are going to simulate down and then !echo hey. I am using Jupyter Notebook hence I have to use ! to use windows commands.

# lets simulate down key and hello world
keyboard.press_and_release("!,e,c,h,o,space,h,e,y")
Enter fullscreen mode Exit fullscreen mode 
# !ECHO HEY
Enter fullscreen mode Exit fullscreen mode

We can even use keys like control.

# lets simulate down key and hello world
keyboard.press_and_release("h,e,l,l,o,ctrl+a")
Enter fullscreen mode Exit fullscreen mode

HELLO

The focus is not on the Keyboard package but to play a dino game. We will do something like gesture recognition based on the distance between certain landmarks. So lets define a method to find Euclidean distance.

def euclidean(pt1, pt2):
    d = np.sqrt((pt1[0]-pt2[0])**2+(pt1[1]-pt2[1])**2)
    return d
euclidean((4, 3), (0, 0))
Enter fullscreen mode Exit fullscreen mode 
5.0
Enter fullscreen mode Exit fullscreen mode

Writing a Code

Step By Step

It is necessary to view the landmark position before making a gesture assumptions. Please follow the below image.

img
Source: Official Hands Page

  • Start by beginning a camera. 
cam = cv2.VideoCapture(0)
Enter fullscreen mode Exit fullscreen mode
  • Define a frame size in our case 520 rows and 720 columns. 
fsize = (520, 720)
Enter fullscreen mode Exit fullscreen mode
  • Take modules drawing_utilities and hands from Mediapipe solutions's. As the name, drawing_utils will draw landmark here and the hands will let us work with detection models. 
mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands
Enter fullscreen mode Exit fullscreen mode
  • Define a variable to count the frame and make a constant to check the events on those frame count. 
check_every = 10
check_cnt = 0
Enter fullscreen mode Exit fullscreen mode
  • Prepare a variable to hold last event name. 
last_event = None
Enter fullscreen mode Exit fullscreen mode
  • Now prepare a Mediapipe Hand object by giving arguments like max_num_hands, min_detection_confidence and so on. As name suggests, max_num_hands is to search up to that number of hands and min_detection_confidence is the minimum confidence threshold value of detection and below which, detected hands are discarded. 
with mp_hands.Hands(
static_image_mode=True,
max_num_hands = 2,
min_detection_confidence=0.6) as hands:
Enter fullscreen mode Exit fullscreen mode
  • Read a Camera frame. 
    while cam.isOpened():
        ret, frame = cam.read()
        if not ret:
            continue
Enter fullscreen mode Exit fullscreen mode
  • Flip the frame to look like selfie camera. 
        frame = cv2.flip(frame, 1)
Enter fullscreen mode Exit fullscreen mode
  • Resize frame to our desired size. 
        frame = cv2.resize(frame, (fsize[1], fsize[0]))
Enter fullscreen mode Exit fullscreen mode
  • Extract width and height of frame. 
        h, w,_ = frame.shape
Enter fullscreen mode Exit fullscreen mode
  • Convert frame from BGR to RGB because Hand object expects image as a RGB format. 
        rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
Enter fullscreen mode Exit fullscreen mode
  • Pass the RGB image ot process module of Hand object to get the result. 
        res = hands.process(rgb)
Enter fullscreen mode Exit fullscreen mode
  • Now for each hand, we will be extracting landmarks of fingers. Like index finger's tip, dip, middle and so on. There are overall 20 landmarks for each hand. After extracting, we need to convert it back to pixel coordinate world. 
        if res.multi_hand_landmarks:
            for hand_landmarks in res.multi_hand_landmarks:

                index_tip = mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_TIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_TIP].y, 
                    w, h)

                thumb_tip = mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.THUMB_TIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.THUMB_TIP].y, 
                    w, h)

                middle_tip = mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.MIDDLE_FINGER_TIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.MIDDLE_FINGER_TIP].y, 
                    w, h)
Enter fullscreen mode Exit fullscreen mode
  • Now if the current count of frame is equal to the value we defined earlier, then check for events. 
                if index_tip is not None:
                    if check_cnt==check_every:
Enter fullscreen mode Exit fullscreen mode
  • If the distance between index finger's tip and middle finger's tip is less than 60 then consider that space is pressed. i.e touch index finger and middle finger for Jump. The value 60 will be relative to the frame size. Else, if last event is also Jump, then set last event to none. 
                        if index_tip is not None and middle_tip is not None:

                            if euclidean(index_tip, middle_tip)<40: 
                                last_event = "jump"
                            else:
                                if last_event=="jump":
                                    last_event=None
Enter fullscreen mode Exit fullscreen mode
  • If the distance between index pip, and thumb tip is less than 60 then consider that down key is pressed. i.e move thumb near to the bottom of index finger for duck. Else if last event is also duck, then set last event to none. 
                        if thumb_tip is not None and index_tip is not None:
                            if euclidean(thumb_tip, index_pip) < 60: # 60 should be relative to height/width of frame
                                last_event = "duck"
                            else:
                                if last_event=="duck":
                                    last_event=None
Enter fullscreen mode Exit fullscreen mode
  • After checking all events, set frame count to 0. 
                    check_cnt=0
Enter fullscreen mode Exit fullscreen mode
  • * Finally, if current frame count has been reseted then apply the event. And increase the frame count. 
            if check_cnt==0:
                if last_event=="jump":
                    keyboard.press_and_release("space")
                elif last_event=="duck":
                    keyboard.press("down")
                else:
                    keyboard.release("down")
                print(last_event)


            check_cnt+=1
Enter fullscreen mode Exit fullscreen mode
  • Draw each landmarks. 
            mp_drawing.draw_landmarks(frame, hand_landmarks, mp_hands.HAND_CONNECTIONS)```
{% endraw %}

* Show the frame.
{% raw %}

```python
        cv2.imshow("Controller Window", frame)
Enter fullscreen mode Exit fullscreen mode
  • If Escape is pressed then close. 
        if cv2.waitKey(1)&0xFF == 27:
            break
Enter fullscreen mode Exit fullscreen mode

Final Code 

cam = cv2.VideoCapture(0)
fsize = (520, 720)

last_event = None
check_cnt = 0
check_every = 5

mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands




with mp_hands.Hands(
static_image_mode=True,
max_num_hands = 1,
min_detection_confidence=0.6) as hands:
    while cam.isOpened():
        ret, frame = cam.read()
        if not ret:
            continue
        frame = cv2.flip(frame, 1)
        frame = cv2.resize(frame, (fsize[1], fsize[0]))

        h, w,_ = frame.shape

        rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        rgb.flags.writeable = False

        res = hands.process(rgb)
        #cv2.imshow("roi", roi)
        rgb.flags.writeable = True


        if res.multi_hand_landmarks:
            for hand_landmarks in res.multi_hand_landmarks:

                index_dip = mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_DIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_DIP].y, 
                    w, h)

                index_tip = mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_TIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_TIP].y, 
                    w, h)

                index_pip = np.array(mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_PIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_PIP].y, 
                    w, h))

                thumb_tip = mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.THUMB_TIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.THUMB_TIP].y, 
                    w, h)

                middle_tip = mp_drawing._normalized_to_pixel_coordinates(
                    hand_landmarks.landmark[mp_hands.HandLandmark.MIDDLE_FINGER_TIP].x, 
                    hand_landmarks.landmark[mp_hands.HandLandmark.MIDDLE_FINGER_TIP].y, 
                    w, h)

                if index_tip is not None:
                    if check_cnt==check_every:
                        if index_tip is not None and middle_tip is not None:


                            if euclidean(index_tip, middle_tip)<40: # 60 should be relative to the height of frame
                                last_event = "jump"
                            else:
                                if last_event=="jump":
                                    last_event=None

                        if thumb_tip is not None and index_tip is not None:
                            print(euclidean(index_tip, middle_tip))
                            if euclidean(thumb_tip, index_tip) < 60:
                                last_event="duck"
                            else:
                                if last_event == "duck":
                                    last_event = None
                        check_cnt=0

                if check_cnt==0:
                    if last_event=="jump":
                        keyboard.press_and_release("space")
                    elif last_event=="duck":
                        keyboard.press("down")
                    else:
                        keyboard.release("down")
                    print(last_event)

                check_cnt+=1

                mp_drawing.draw_landmarks(frame, hand_landmarks, mp_hands.HAND_CONNECTIONS)
        cv2.imshow("Controller Window", frame)

        if cv2.waitKey(1)&0xFF == 27:
            break
cam.release()
cv2.destroyAllWindows()
Enter fullscreen mode Exit fullscreen mode

Finally

This is the end of our blog and I hope you learned something valuable from here. Please let me know if you found any problems or errors. There is a video version of this blog and you can watch this on YouTube too.

💖 💪 🙅 🚩
qviper
Viper

Posted on July 19, 2021

Join Our Newsletter. No Spam, Only the good stuff.

Sign up to receive the latest update from our blog.

Related

How to Run Samurai on Google Colab
googlecolab How to Run Samurai on Google Colab

November 26, 2024

These Are the Best Programming Languages for AI Development.
ai These Are the Best Programming Languages for AI Development.

November 21, 2024

Introducing InsightfulAI: Public Alpha API for Simplified Machine Learning
python Introducing InsightfulAI: Public Alpha API for Simplified Machine Learning

November 14, 2024

My First Data Analysis Project
beginners My First Data Analysis Project

November 6, 2024

🎉 Exciting Research Throwback! 🎉
computerscience 🎉 Exciting Research Throwback! 🎉

November 18, 2024