如何基于摄像头指甲检测在指定手部Landmark位置叠加静态图片

当前效果 #

当前已实现手部关键点的圆形标记，仅显示拇指尖、食指尖、中指尖、无名指尖、小指尖这五个关键点，对应的连接配置为：
handConnection = [(4, 4), (8, 8), (12, 12), (16, 16), (20, 20)]

需求说明 #

希望将当前的圆形标记替换为静态图片，为五个关键点分别设置不同的图片，实现每个指尖位置显示对应图片的效果（类似美甲贴纸覆盖在指尖）。

当前完整代码 #

import mediapipe as mp
import cv2
import cvzone

import numpy as np
import uuid
import os

mp_drawing = mp.solutions.drawing_utils
drawSpecific = mp.solutions.hands
mp_hands = mp.solutions.hands
handConnection = [(4, 4), (8, 8), (12, 12), (16, 16), (20, 20)]

cap = cv2.VideoCapture(0)
(ret, image) = cap.read()
imgFront = cv2.imread("poki.png", cv2.IMREAD_UNCHANGED)
imgFront = cv2.resize(imgFront, (0, 0), None, 0.3, 0.3)

hf, wf, cf = imgFront.shape
hb, wb, cb = image.shape

fpsReader = cvzone.FPS()
def main():
    hands = mp_hands.Hands(min_detection_confidence=0.7,
                           min_tracking_confidence=0.7)
    hand_landmark_drawing_spec = mp_drawing.DrawingSpec(color=(121, 22,
            6), thickness=5, circle_radius=0)
    hand_connection_drawing_spec = mp_drawing.DrawingSpec(color=(250,
            44, 90), thickness=15, circle_radius=15)

    while cap.isOpened():

        (ret, image) = cap.read()
        image = cv2.flip(image, 1)
        image = cv2.cvtColor(cv2.flip(image, 1), cv2.COLOR_BGR2RGB)
        (image_height, image_width, _) = image.shape

        imgResult = cvzone.overlayPNG(image, imgFront, [0, hb - hf])
        _, imgResult = fpsReader.update(imgResult)

        results_hand = hands.process(image)
        image.flags.writeable = True
        image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)

        if results_hand.multi_hand_landmarks:
            for (idx, hand_landmarks) in \
                enumerate(results_hand.multi_hand_landmarks):
                color = (0xFF, 0, 0)

                center_coordinates1 = \
                    (hand_landmarks.landmark[drawSpecific.HandLandmark.THUMB_TIP].x
                     * image_width,
                     hand_landmarks.landmark[drawSpecific.HandLandmark.THUMB_TIP].y
                     * image_height)
                center_coordinates2 = \
                    (hand_landmarks.landmark[drawSpecific.HandLandmark.INDEX_FINGER_TIP].x
                     * image_width,
                     hand_landmarks.landmark[drawSpecific.HandLandmark.INDEX_FINGER_TIP].y
                     * image_height)
                center_coordinates3 = \
                    (hand_landmarks.landmark[drawSpecific.HandLandmark.MIDDLE_FINGER_TIP].x
                     * image_width,
                     hand_landmarks.landmark[drawSpecific.HandLandmark.MIDDLE_FINGER_TIP].y
                     * image_height)
                center_coordinates4 = \
                    (hand_landmarks.landmark[drawSpecific.HandLandmark.RING_FINGER_TIP].x
                     * image_width,
                     hand_landmarks.landmark[drawSpecific.HandLandmark.RING_FINGER_TIP].y
                     * image_height)
                center_coordinates5 = \
                    (hand_landmarks.landmark[drawSpecific.HandLandmark.PINKY_TIP].x
                     * image_width,
                     hand_landmarks.landmark[drawSpecific.HandLandmark.PINKY_TIP].y
                     * image_height)

                img2 = cv2.imread("Rafath Auvee.png")
                img2 = cv2.resize(img2, (100,100))


                cv2.circle(image, (int(center_coordinates1[0]),
                           int(center_coordinates1[1])), 15, color, 5)
                cv2.circle(image, (int(center_coordinates2[0]),
                           int(center_coordinates2[1])), 15, color, 5)
                cv2.circle(image, (int(center_coordinates3[0]),
                           int(center_coordinates3[1])), 15, color, 5)
                cv2.circle(image, (int(center_coordinates4[0]),
                           int(center_coordinates4[1])), 15, color, 5)
                cv2.circle(image, (int(center_coordinates5[0]),
                           int(center_coordinates5[1])), 15, color, 5)


        cv2.imshow('Nail Tracking', imgResult)
        if cv2.waitKey(10) & 0xFF == ord('q'):
            break

    hands.close()
    cap.release()


main()

待解决问题 #

如何将代码中当前绘制的五个圆形标记替换为手动导入的五张不同静态图片，让图片精准覆盖在对应的指尖关键点位置？

内容的提问来源于stack exchange，提问作者Auvee