python爬取B站视频

参考：https://cloud.tencent.com/developer/article/1768680

参考的代码有点问题，请求头需要修改，上代码：

import requests
import re  # 正则表达式
import pprint
import json
from moviepy.editor import AudioFileClip, VideoFileClip
from bs4 import BeautifulSoup as bs

headers = {
    # 防盗链 告诉服务器 我们请求的url网址是从哪里跳转过来的
    'referer': 'https://www.bilibili.com/a',
    'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/86.0.4240.198 Safari/537.36'
}

def send_request(url):
    response = requests.get(url=url, headers=headers)
    return response

def get_video_data(html_data):
    """解析视频数据"""

    # 提取视频的标题
    soup = bs(html_data, 'lxml')
    title = soup.find_all(name='h1',attrs={"class":"video-title special-text-indent"})[0].get_text()
    # print(title)

    # 提取视频对应的json数据
    json_data = re.findall('<script>window\.__playinfo__=(.*?)</script>', html_data)[0]
    # print(json_data)  # json_data 字符串
    json_data = json.loads(json_data)
    pprint.pprint(json_data)

    # 提取音频的url地址
    audio_url = json_data['data']['dash']['audio'][0]['backupUrl'][0]
    print('解析到的音频地址:', audio_url)

    # 提取视频画面的url地址
    video_url = json_data['data']['dash']['video'][0]['backupUrl'][0]
    print('解析到的视频地址:', video_url)

    video_data = [title, audio_url, video_url]
    return video_data

def save_data(file_name, audio_url, video_url):
    # 请求数据
    print('正在请求音频数据')
    audio_data = send_request(audio_url).content
    print('正在请求视频数据')
    video_data = send_request(video_url).content
    with open(file_name + '.mp3', mode='wb') as f:
        f.write(audio_data)
        print('正在保存音频数据')
    with open(file_name + '.mp4', mode='wb') as f:
        f.write(video_data)
        print('正在保存视频数据')

def merge_data(video_name):
    print('视频合成开始:', video_name)
    audioclip = AudioFileClip(video_name+'.mp3')
    videoclip = VideoFileClip(video_name+'.mp4')
    # 3.获取视频和音频的时长
    video_time = videoclip.duration
    audio_time = audioclip.duration
    # 4.对视频或者音频进行裁剪
    if video_time > audio_time:
        # 视频时长>音频时长，对视频进行截取
        videoclip_new = videoclip.subclip(0, audio_time)
        audioclip_new = audioclip
    else:
        # 音频时长>视频时长，对音频进行截取
        videoclip_new = videoclip
        audioclip_new = audioclip.subclip(0, video_time)
    # 5.视频中加入音频
    video_with_new_audio = videoclip_new.set_audio(audioclip_new)
    # 6.写入到新的视频文件中
    video_with_new_audio.write_videofile("output.mp4",
                                         codec='libx264',
                                         audio_codec='aac',
                                         temp_audiofile='temp-video.m4a',
                                         remove_temp=True
                                         )
    print('视频合成结束:', video_name)


url = 'https://www.bilibili.com/video/BV1bK421a7qG/?spm_id_from=333.1007.tianma.6-4-22.click'
response = send_request(url)
response.encoding = requests.utils.get_encodings_from_content(response.text)[0]
html_data = response.text
video_data = get_video_data(html_data)
save_data(video_data[0], video_data[1], video_data[2])
merge_data(video_data[0])

效果

小姐姐挺靓，就是左下角有水印，想办法去除水印，参考：python实战之去除视频水印&字幕_python 去除视频水印-CSDN博客

import os
import sys
import cv2
import numpy
from moviepy import editor
 
TEMP_VIDEO = 'temp.mp4'
 
 
class WatermarkRemover():
 
    def __init__(self, video_path, output, threshold: int, kernel_size: int):
        self.threshold = threshold  # 阈值分割所用阈值
        self.kernel_size = kernel_size  # 膨胀运算核尺寸
        self.video_path = video_path
        self.output = output
 
 
    #根据用户手动选择的ROI（Region of Interest，感兴趣区域）框选水印或字幕位置。
    def select_roi(self, img: numpy.ndarray, hint: str) -> list:
        '''
    框选水印或字幕位置，SPACE或ENTER键退出
    :param img: 显示图片
    :return: 框选区域坐标
    '''
        COFF = 0.7
        w, h = int(COFF * img.shape[1]), int(COFF * img.shape[0])
        resize_img = cv2.resize(img, (w, h))
        roi = cv2.selectROI(hint, resize_img, False, False)
        cv2.destroyAllWindows()
        watermark_roi = [int(roi[0] / COFF), int(roi[1] / COFF), int(roi[2] / COFF), int(roi[3] / COFF)]
        return watermark_roi
 
 
    #对输入的蒙版进行膨胀运算，扩大蒙版的范围
    def dilate_mask(self, mask: numpy.ndarray) -> numpy.ndarray:
 
        '''
    对蒙版进行膨胀运算
    :param mask: 蒙版图片
    :return: 膨胀处理后蒙版
    '''
        kernel = numpy.ones((self.kernel_size, self.kernel_size), numpy.uint8)
        mask = cv2.dilate(mask, kernel)
        return mask
    
    #根据手动选择的ROI区域，在单帧图像中生成水印或字幕的蒙版。
    def generate_single_mask(self, img: numpy.ndarray, roi: list, threshold: int) -> numpy.ndarray:
        '''
    通过手动选择的ROI区域生成单帧图像的水印蒙版
    :param img: 单帧图像
    :param roi: 手动选择区域坐标
    :param threshold: 二值化阈值
    :return: 水印蒙版
    '''
        # 区域无效，程序退出
        if len(roi) != 4:
            print('NULL ROI!')
            sys.exit()
 
        # 复制单帧灰度图像ROI内像素点
        roi_img = numpy.zeros((img.shape[0], img.shape[1]), numpy.uint8)
        start_x, end_x = int(roi[1]), int(roi[1] + roi[3])
        start_y, end_y = int(roi[0]), int(roi[0] + roi[2])
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        roi_img[start_x:end_x, start_y:end_y] = gray[start_x:end_x, start_y:end_y]
 
        # 阈值分割
        _, mask = cv2.threshold(roi_img, threshold, 255, cv2.THRESH_BINARY)
        return mask
 
    #通过截取视频中多帧图像生成多张水印蒙版，并通过逻辑与计算生成最终的水印蒙版
    def generate_watermark_mask(self, video_path: str) -> numpy.ndarray:
        '''
    截取视频中多帧图像生成多张水印蒙版，通过逻辑与计算生成最终水印蒙版
    :param video_path: 视频文件路径
    :return: 水印蒙版
    '''
        video = cv2.VideoCapture(video_path)
        success, frame = video.read()
        roi = self.select_roi(frame, 'select watermark ROI')
        mask = numpy.ones((frame.shape[0], frame.shape[1]), numpy.uint8)
        mask.fill(255)
 
        step = video.get(cv2.CAP_PROP_FRAME_COUNT) // 5
        index = 0
        while success:
            if index % step == 0:
                mask = cv2.bitwise_and(mask, self.generate_single_mask(frame, roi, self.threshold))
            success, frame = video.read()
            index += 1
        video.release()
 
        return self.dilate_mask(mask)
 
    #根据手动选择的ROI区域，在单帧图像中生成字幕的蒙版。
    def generate_subtitle_mask(self, frame: numpy.ndarray, roi: list) -> numpy.ndarray:
        '''
    通过手动选择ROI区域生成单帧图像字幕蒙版
    :param frame: 单帧图像
    :param roi: 手动选择区域坐标
    :return: 字幕蒙版
    '''
        mask = self.generate_single_mask(frame, [0, roi[1], frame.shape[1], roi[3]], self.threshold)  # 仅使用ROI横坐标区域
        return self.dilate_mask(mask)
 
    def inpaint_image(self, img: numpy.ndarray, mask: numpy.ndarray) -> numpy.ndarray:
        '''
    修复图像
    :param img: 单帧图像
    :parma mask: 蒙版
    :return: 修复后图像
    '''
        telea = cv2.inpaint(img, mask, 1, cv2.INPAINT_TELEA)
        return telea
 
 
    def merge_audio(self, input_path: str, output_path: str, temp_path: str):
        '''
    合并音频与处理后视频
    :param input_path: 原视频文件路径
    :param output_path: 封装音视频后文件路径
    :param temp_path: 无声视频文件路径
    '''
        with editor.VideoFileClip(input_path) as video:
            audio = video.audio
            with editor.VideoFileClip(temp_path) as opencv_video:
                clip = opencv_video.set_audio(audio)
                clip.to_videofile(output_path)
 
    def remove_video_watermark(self):
        '''
    去除视频水印
    '''
        if not os.path.exists(self.output):
            os.makedirs(self.output)
 
        filenames = [os.path.join(self.video_path, i) for i in os.listdir(self.video_path)]
        mask = None
 
        for i, name in enumerate(filenames):
            if i == 0:
                # 生成水印蒙版
                mask = self.generate_watermark_mask(name)
 
            # 创建待写入文件对象
            video = cv2.VideoCapture(name)
            fps = video.get(cv2.CAP_PROP_FPS)
            size = (int(video.get(cv2.CAP_PROP_FRAME_WIDTH)), int(video.get(cv2.CAP_PROP_FRAME_HEIGHT)))
            video_writer = cv2.VideoWriter(TEMP_VIDEO, cv2.VideoWriter_fourcc(*'mp4v'), fps, size)
 
            # 逐帧处理图像
            success, frame = video.read()
 
            while success:
                frame = self.inpaint_image(frame, mask)
                video_writer.write(frame)
                success, frame = video.read()
 
            video.release()
            video_writer.release()
 
            # 封装视频
            (_, filename) = os.path.split(name)
            output_path = os.path.join(self.output, filename.split('.')[0] + '_no_watermark.mp4')  # 输出文件路径
            self.merge_audio(name, output_path, TEMP_VIDEO)
 
    if os.path.exists(TEMP_VIDEO):
        os.remove(TEMP_VIDEO)
 
    def remove_video_subtitle(self):
        '''去除视频字幕'''
        if not os.path.exists(self.output):
            os.makedirs(self.output)
 
        filenames = [os.path.join(self.video_path, i) for i in os.listdir(self.video_path)]
        roi = []
 
        for i, name in enumerate(filenames):
            # 创建待写入文件对象
            video = cv2.VideoCapture(name)
            fps = video.get(cv2.CAP_PROP_FPS)
            size = (int(video.get(cv2.CAP_PROP_FRAME_WIDTH)), int(video.get(cv2.CAP_PROP_FRAME_HEIGHT)))
            video_writer = cv2.VideoWriter(TEMP_VIDEO, cv2.VideoWriter_fourcc(*'mp4v'), fps, size)
 
            # 逐帧处理图像
            success, frame = video.read()
            if i == 0:
                roi = self.select_roi(frame, 'select subtitle ROI')
 
            while success:
                mask = self.generate_subtitle_mask(frame, roi)
                frame = self.inpaint_image(frame, mask)
                video_writer.write(frame)
                success, frame = video.read()
 
            video.release()
            video_writer.release()
 
            # 封装视频
            (_, filename) = os.path.split(name)
            output_path = os.path.join(OUTPUT_PATH, filename.split('.')[0] + '_no_sub.mp4')  # 输出文件路径
            self.merge_audio(name, output_path, TEMP_VIDEO)
 
        if os.path.exists(TEMP_VIDEO):
            os.remove(TEMP_VIDEO)
 
 # 去水印
video_path = 'video'
output_path = 'output'
remover = WatermarkRemover(video_path,output_path,threshold=80, kernel_size=5)
remover.remove_video_watermark()   
#去字幕
# remover = WatermarkRemover(video_path,output_path,threshold=80, kernel_size=5)
# remover.remove_video_subtitle()

效果一般吧：