pixellib防弹幕遮挡

注意：请使用conda创建新环境安装，否则可能会出现各种各样的bug

实现思路：

1. 使用pixellib把视频的人使用实例分割，存储为灰度图，如果有多个人则合并成一张图。每一帧存储为一张遮住图

   import cv2
   import numpy as np
   from pixellib.instance import instance_segmentation
   
   class VideoProcessor:
       def __init__(self, videoFile):
           self.videoFile = videoFile
           pass
   
       def video2mask(self):
           instance = instance_segmentation()
           instance.load_model('./data/mask_rcnn_coco.h5')
           target = instance.select_target_classes(person=True)
   
           cap = cv2.VideoCapture(self.videoFile)
           frame_index = 0
           while True:
               ret, frame = cap.read()
               if not ret:
                   print("视频处理完毕")
                   break
   
               target_class = instance.select_target_classes(person=True)
               result, output = instance.segmentFrame(frame=frame, show_bboxes=True, segment_target_classes=target_class)
               person_num = len(result['class_ids'])
               if person_num > 0:
                   print('第' + str(frame_index) + '帧，检测到' + str(person_num) + '个人')
                   masks = result['masks']
   
                   black_mask = np.zeros(frame.shape[:2], dtype=np.uint8)
   
                   for i in range(person_num):
                       print('第' + str(frame_index) + '帧，检测到第' + str(i + 1) + '个人的脸部')
                       black_mask = np.where(masks[:, :, i], 255, black_mask)
   
                   mask_file_name = './masks/' + str(frame_index) + '.jpg'
                   cv2.imwrite(mask_file_name, black_mask)
   
               frame_index = frame_index + 1
   
               cv2.imshow('frame', output)
               if cv2.waitKey(10) & 0xFF == ord('q'):
                   break
           cap.release()
           cv2.destroyAllWindows()
           pass
   
   
   vp = VideoProcessor('./data/video.mp4')
   vp.video2mask()

   最后输出如下结果，中间空白部分用来判断是否需要透明

   

2. 将视频读入，将遮罩图读入，通过黑白来判断这个地方是否需要展示弹幕
3. 创建蒙版层，将透明度设置为0
4. 将文字创建在蒙版上
5. 用np.where判断遮住图的(0,0)等是否为255，如果是说明这个地方是人，那么这个像素的alpha通道就显示为0透明，否则就显示蒙版的像素
   注意：创建遮住代表全部透明，嵌入文字后文字地方不透明，最后再修改整个遮住透明度，因为这个时候文字已经嵌入，和遮住合并为一体了

from random import randint

import cv2
import numpy as np
from PIL import Image, ImageDraw, ImageFont
import matplotlib.pyplot as plt

frame_index = 0
width = 0
height = 0
font_size = 40
lane_center = font_size / 2
lane_num = 0
color = (255, 0, 255, 0)
cap = cv2.VideoCapture('data/video.mp4')
while True:
    # 读取图片
    ret, frame = cap.read()
    # 设置每帧的宽高，和泳道lane数量
    if width < 1 or bool(height) < 1 or bool(lane_num) < 1:
        width = frame.shape[1]
        height = frame.shape[0]
        print(frame.shape[0])

        lane_num = height / font_size

    if not ret:
        print("视频处理完毕")
        break
    # 获取遮罩图片
    mask_bg = cv2.imread("./masks/" + str(frame_index) + ".jpg", cv2.IMREAD_GRAYSCALE)
    print(mask_bg.shape)

    # 创建蒙版
    bg = Image.new("RGBA", (width, height), (255, 0, 255, 0))

    # 设置字体
    font_style = ImageFont.truetype("./MSYH.TTC", 80, encoding="utf-8")
    # 写字
    draw_bg = ImageDraw.Draw(bg)
    draw_bg.text((0, font_size * randint(1, lane_num) - 20), "当前第:" + str(frame_index), fill=(0, 0, 0, 255), font=font_style)
    rgba_bg = np.array(bg)
    # 设置透明区域
    rgba_bg[:, :, 3] = np.where(mask_bg == 255, 0, rgba_bg[:, :, 3])
    # 转换倒pil并粘贴弹幕蒙版
    frame_image = Image.fromarray(cv2.cvtColor(frame,cv2.COLOR_BGR2RGBA))
    bg = Image.fromarray(rgba_bg)
    # frame_image.paste(bg)
    re = Image.alpha_composite(frame_image, bg)
    # cv2.imshow("e", frame_image)
    plt.imshow(re)
    plt.show()
    frame_index+=1