Hand Gesture Recognition Using OpenCV and Python: Video Capture, Skin Detection, and Contour Processing

The tutorial walks through creating a simple hand‑gesture recognition project that can be followed by beginners, using only OpenCV, basic Python syntax, and fundamental image‑processing concepts.

Video Capture : The script starts by opening a video file (or webcam) with cv2.VideoCapture("C:/Users/lenovo/Videos/1.mp4") , reads frames in a loop, resizes them, draws a rectangle to define the region of interest, and extracts that ROI for further processing.

cap = cv2.VideoCapture("C:/Users/lenovo/Videos/1.mp4")  # read file
while True:
    ret, frame = cap.read()
    src = cv2.resize(frame, (400, 350), interpolation=cv2.INTER_CUBIC)
    cv2.rectangle(src, (90, 60), (300, 300), (0, 255, 0))
    roi = src[60:300, 90:300]
    ...
    if cv2.waitKey(50) & 0xFF == ord('q'):
        break
cap.release()
cv2.destroyAllWindows()

Skin Detection : The ROI is converted to the YCrCb color space to reduce the influence of illumination, then the Cr channel is blurred and thresholded with Otsu's method to obtain a binary skin mask. The mask is applied to the original ROI using cv2.bitwise_and .

def A(img):
    YCrCb = cv2.cvtColor(img, cv2.COLOR_BGR2YCR_CB)
    y, cr, cb = cv2.split(YCrCb)
    cr1 = cv2.GaussianBlur(cr, (5, 5), 0)
    _, skin = cv2.threshold(cr1, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
    res = cv2.bitwise_and(img, img, mask=skin)
    return res

Contour Processing : After converting the skin‑detected image to grayscale and applying a Laplacian filter, contours are found with cv2.findContours . The contours are sorted by area and the largest one (assumed to be the hand) is drawn on a white canvas.

def B(img):
    h = cv2.findContours(img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
    contour = h[0]
    contour = sorted(contour, key=cv2.contourArea, reverse=True)
    bg = np.ones(dst.shape, np.uint8) * 255
    ret = cv2.drawContours(bg, contour[0], -1, (0, 0, 0), 3)
    return ret

The main loop applies the skin detection function A to the ROI, converts the result to grayscale, computes the Laplacian, and then feeds it to the contour function B . The intermediate results are displayed in separate windows titled "0" (ROI), "2" (contour), and the original video stream.

All code snippets are provided in a single block at the end of the article, allowing readers to copy and run the program directly.