Python Tutorial: Download Bilibili Video, Extract Frames, Perform Human Segmentation with Baidu AI, Generate Word Cloud, and Compose Final Video

This guide walks through a complete Python workflow for processing a Bilibili dance video, including downloading, frame extraction, AI‑based human segmentation, word‑cloud creation, and final video composition with audio.

1. Install required libraries

<code>import os</code><code>import time</code><code>libs = {"lxml","requests","pandas","numpy","you-get","opencv-python","pandas","fake_useragent","matplotlib","moviepy"}</code><code>try:</code><code>    for lib in libs:</code><code>        os.system(f"pip3 install -i https://pypi.doubanio.com/simple/ {lib}")</code><code>        print(lib+"下载成功")</code><code>except:</code><code>    print("下载失败")</code>

2. Import modules

<code>import os</code><code>import re</code><code>import cv2</code><code>import jieba</code><code>import requests</code><code>import moviepy</code><code>import pandas as pd</code><code>import numpy as np</code><code>from PIL import Image</code><code>from lxml import etree</code><code>from wordcloud import WordCloud</code><code>import matplotlib.pyplot as plt</code><code>from fake_useragent import UserAgent</code>

3. Download the video using you-get (the original Bilibili URL is provided in the article).

4. Split video into frames with OpenCV:

<code># -*- coding:utf-8 -*-</code><code># @Author : 北山啦</code><code># @File : 视频分割.py</code><code>import cv2</code><code>cap = cv2.VideoCapture(r"无价之姐~让我乘风破浪~~~.flv")</code><code>while 1:</code><code>    ret, frame = cap.read()</code><code>    if ret:</code><code>        cv2.imwrite(f"./pictures/img_{num}.jpg", frame)</code><code>    else:</code><code>        break</code><code>cap.release()   # 释放资源</code>

5. Human segmentation using Baidu AI (Python SDK):

<code>import cv2</code><code>import base64</code><code>import numpy as np</code><code>import os</code><code>from aip import AipBodyAnalysis</code><code>APP_ID = '******'</code><code>API_KEY = '*******************'</code><code>SECRET_KEY = '********************'</code><code>client = AipBodyAnalysis(APP_ID, API_KEY, SECRET_KEY)</code><code>path = './mask_img/'</code><code>img_files = os.listdir('./pictures')</code><code>for num in range(1, len(img_files)+1):</code><code>    img = f'./pictures/img_{num}.jpg'</code><code>    img1 = cv2.imread(img)</code><code>    height, width, _ = img1.shape</code><code>    with open(img, 'rb') as fp:</code><code>        img_info = fp.read()</code><code>    seg_res = client.bodySeg(img_info)</code><code>    labelmap = base64.b64decode(seg_res['labelmap'])</code><code>    nparr = np.frombuffer(labelmap, np.uint8)</code><code>    labelimg = cv2.imdecode(nparr, 1)</code><code>    labelimg = cv2.resize(labelimg, (width, height), interpolation=cv2.INTER_NEAREST)</code><code>    new_img = np.where(labelimg == 1, 255, labelimg)</code><code>    mask_name = path + f'mask_{num}.png'</code><code>    cv2.imwrite(mask_name, new_img)</code><code>    print(f'======== 第{num}张图像分割完成 ========')</code>

6. Generate a word‑cloud animation from the extracted comments:

<code># -*- coding:utf-8 -*-</code><code># @Author : 北山啦</code><code># @File : 跳舞词云图生成.py</code><code>from wordcloud import WordCloud</code><code>import collections</code><code>import jieba</code><code>import re</code><code>from PIL import Image</code><code>import matplotlib.pyplot as plt</code><code>import numpy as np</code><code>with open('barrages.txt') as f:</code><code>    data = f.read()</code><code>jieba.load_userdict("./词云图//add.txt")</code><code>new_data = "/".join(re.findall('[\u4e00-\u9fa5]+', data, re.S))</code><code>seg_list_exact = jieba.cut(new_data, cut_all=True)</code><code>result_list = []</code><code>with open('./词云图/stoplist.txt', encoding='utf-8') as f:</code><code>    stop_words = set(f.read().split('\n'))</code><code>for word in seg_list_exact:</code><code>    if word not in stop_words and len(word) > 1:</code><code>        result_list.append(word)</code><code>word_counts = collections.Counter(result_list)</code><code>mask_ = 255 - np.array(Image.open('./mask_img/mask_1.png'))</code><code>my_cloud = WordCloud(background_color='black', mask=mask_, mode='RGBA', max_words=500, font_path='simhei.ttf').generate_from_frequencies(word_counts)</code><code>plt.figure(figsize=(8,5), dpi=200)</code><code>plt.imshow(my_cloud)</code><code>plt.axis('off')</code><code>my_cloud.to_file('./wordcloud/wordcloud_1.png')</code><code>print(f'======== 第1张词云图生成 ========')</code>

7. Compose the final video from the word‑cloud images using OpenCV:

<code># -*- coding:utf-8 -*-</code><code># @Author : 北山啦</code><code># @File : 跳舞词云图生成.py</code><code>import cv2</code><code>import os</code><code>video_dir = 'result.mp4'</code><code>fps = 30</code><code>img_size = (1920, 1080)</code><code>fourcc = cv2.VideoWriter_fourcc('M','P','4','V')</code><code>videoWriter = cv2.VideoWriter(video_dir, fourcc, fps, img_size)</code><code>img_files = os.listdir('./wordcloud')</code><code>for i in range(88, 888):</code><code>    img_path = f'./wordcloud/wordcloud_{i}.png'</code><code>    frame = cv2.imread(img_path)</code><code>    frame = cv2.resize(frame, img_size)</code><code>    videoWriter.write(frame)</code><code>    print(f'======== 按照视频顺序第{i}张图片合进视频 ========')</code><code>videoWriter.release()   # 释放资源</code>

Finally, add background music with moviepy :

<code># -*- coding:utf-8 -*-</code><code># @Author : 北山啦</code><code>import moviepy.editor as mpy</code><code>my_clip = mpy.VideoFileClip('result.mp4')</code><code>audio_background = mpy.AudioFileClip('song.mp3').subclip(0,25)</code><code>final_clip = my_clip.set_audio(audio_background)</code><code>final_clip.write_videofile('final_video.mp4')</code>

The resulting video combines the dance footage, AI‑segmented human masks, and animated word clouds, all synchronized with a background soundtrack.