首先给出展示结果,大体就是检测工业板子是否出现。采取检测的方法比较简单,用的OpenCV的模板检测。
-
opencv读取视频
-
将视频分割为帧
-
对每一帧进行处理(opencv模板匹配)
-
在将此帧写入pipe管道
-
利用ffmpeg进行推流直播
中间遇到的问题
在处理本地视频时,并没有延时卡顿的情况。但对实时视频流的时候,出现了卡顿延时的效果。在一顿度娘操作之后,采取了多线程的方法。
opencv读取视频
def run_opencv_camera():
video_stream_path = 0
cap = cv2.VideoCapture(video_stream_path)
while cap.isOpened():
is_opened, frame = cap.read()
cv2.imshow('frame', frame)
cv2.waitKey(1)
cap.release()
OpenCV模板匹配
模板匹配就是在一幅图像中寻找一个特定目标的方法之一,这种方法的原理非常简单,遍历图像中每一个可能的位置,比较各处与模板是否相似,当相似度足够高时,就认为找到了目标。
def template_match(img_rgb):
img_gray = cv2.cvtColor(img_rgb, cv2.COLOR_BGR2GRAY)
res = cv2.matchTemplate(img_gray, template, cv2.TM_CCOEFF_NORMED)
threshold = 0.8
loc = np.where(res >= threshold)
if len(loc[0]):
cv2.rectangle(img_rgb, (155, 515), (1810, 820), (0, 0, 255), 3)
cv2.putText(img_rgb, category, (240, 600), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, Confidence, (240, 640), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, Precision, (240, 680), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, product_yield, (240, 720), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, result, (240, 780), cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 255, 0), 5)
return img_rgb
FFmpeg推流
- 在Ubuntu 14 上安装 Nginx-RTMP 流媒体服务器
https://www.cnblogs.com/cocoajin/p/4353767.html
import subprocess as sp
rtmpUrl = ""
camera_path = ""
cap = cv.VideoCapture(camera_path)
fps = int(cap.get(cv.CAP_PROP_FPS))
width = int(cap.get(cv.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv.CAP_PROP_FRAME_HEIGHT))
command = ['ffmpeg',
'-y',
'-f', 'rawvideo',
'-vcodec','rawvideo',
'-pix_fmt', 'bgr24',
'-s', "{}x{}".format(width, height),
'-r', str(fps),
'-i', '-',
'-c:v', 'libx264',
'-pix_fmt', 'yuv420p',
'-preset', 'ultrafast',
'-f', 'flv',
rtmpUrl]
p = sp.Popen(command, stdin=sp.PIPE)
while(cap.isOpened()):
ret, frame = cap.read()
if not ret:
print("Opening camera is failed")
break
p.stdin.write(frame.tostring())
- 说明:rtmp是要接受视频的服务器,服务器按照上面所给连接地址即可。
多线程处理
- python mutilprocessing多进程编程 https://blog.csdn.net/jeffery0207/article/details/82958520
def image_put(q):
cap = cv2.VideoCapture("./new.mp4")
if cap.isOpened():
print('success')
else:
print('faild')
while True:
q.put(cap.read()[1])
q.get() if q.qsize() > 1 else time.sleep(0.01)
def image_get(q):
while True:
frame = q.get()
frame = template_match(frame)
cv2.imshow("frame", frame)
cv2.waitKey(0)
def run_single_camera():
mp.set_start_method(method='spawn')
queue = mp.Queue(maxsize=2)
processes = [mp.Process(target=image_put, args=(queue, )),
mp.Process(target=image_get, args=(queue, ))]
[process.start() for process in processes]
[process.join() for process in processes]
def run():
run_single_camera()
- 说明:使用Python3自带的多线程模块mutilprocessing模块,创建一个队列,线程A从通过rstp协议从视频流中读取出每一帧,并放入队列中,线程B从队列中将图片取出,处理后进行显示。线程A如果发现队列里有两张图片,即线程B的读取速度跟不上线程A,那么线程A主动将队列里面的旧图片删掉,换新图片。
全部代码展示
import time
import multiprocessing as mp
import numpy as np
import random
import subprocess as sp
import cv2
import os
template_path = "./high_img_template.jpg"
category = "Category: board"
var_confidence = (np.random.randint(86, 98)) / 100
Confidence = "Confidence: " + str(var_confidence)
var_precision = round(random.uniform(98, 99), 2)
Precision = "Precision: " + str(var_precision) + "%"
product_yield = "Product Yield: 100%"
result = "Result: perfect"
template = cv2.imread(template_path, 0)
h, w = template.shape[:2]
def template_match(img_rgb):
img_gray = cv2.cvtColor(img_rgb, cv2.COLOR_BGR2GRAY)
res = cv2.matchTemplate(img_gray, template, cv2.TM_CCOEFF_NORMED)
threshold = 0.8
loc = np.where(res >= threshold)
if len(loc[0]):
cv2.rectangle(img_rgb, (155, 515), (1810, 820), (0, 0, 255), 3)
cv2.putText(img_rgb, category, (240, 600), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, Confidence, (240, 640), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, Precision, (240, 680), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, product_yield, (240, 720), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.putText(img_rgb, result, (240, 780), cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 255, 0), 5)
return img_rgb
size = (1920, 1080)
sizeStr = str(size[0]) + 'x' + str(size[1])
fps = 11
hz = int(1000.0 / fps)
print ('size:'+ sizeStr + ' fps:' + str(fps) + ' hz:' + str(hz))
rtmpUrl = 'rtmp://localhost/hls/test'
command = ['ffmpeg',
'-y',
'-f', 'rawvideo',
'-vcodec','rawvideo',
'-pix_fmt', 'bgr24',
'-s', sizeStr,
'-r', str(fps),
'-i', '-',
'-c:v', 'libx264',
'-pix_fmt', 'yuv420p',
'-preset', 'ultrafast',
'-f', 'flv',
rtmpUrl]
pipe = sp.Popen(command, stdin=sp.PIPE)
def image_put(q):
cap = cv2.VideoCapture("./new.mp4")
if cap.isOpened():
print('success')
else:
print('faild')
while True:
q.put(cap.read()[1])
q.get() if q.qsize() > 1 else time.sleep(0.01)
save_path = "./res_imgs"
if os.path.exists(save_path):
os.makedir(save_path)
def image_get(q):
while True:
frame = q.get()
frame = template_match(frame)
cv2.imshow("frame", frame)
cv2.waitKey(0)
def run_single_camera():
mp.set_start_method(method='spawn')
queue = mp.Queue(maxsize=2)
processes = [mp.Process(target=image_put, args=(queue, )),
mp.Process(target=image_get, args=(queue, ))]
[process.start() for process in processes]
[process.join() for process in processes]
def run():
run_single_camera()
if __name__ == '__main__':
run()
- 在Ubuntu 14 上安装 Nginx-RTMP 流媒体服务器:https://www.cnblogs.com/cocoajin/p/4353767.html
- python mutilprocessing多进程编程:https://blog.csdn.net/jeffery0207/article/details/82958520
- ffmpeg 将视频和图片互转化:https://blog.csdn.net/TingiBanDeQu/article/details/53896944
- 基于python2.7的opencv3.3-ffmpeg-rtmp视频处理并推送流直播:https://blog.csdn.net/u014303844/article/details/80394101
- 读取多个(海康\大华)网络摄像头的视频流 (使用opencv-python),解决实时读取延迟问题:https://zhuanlan.zhihu.com/p/38136322
- python利用ffmpeg进行rtmp推流直播:https://zhuanlan.zhihu.com/p/74260950
Python实现推流直播首先给出展示结果,大体就是检测工业板子是否出现。采取检测的方法比较简单,用的OpenCV的模板检测。大体思路opencv读取视频将视频分割为帧对每一帧进行处理(opencv模板匹配)在将此帧写入pipe管道利用ffmpeg进行推流直播中间遇到的问题在处理本地视频时,并没有延时卡顿的情况。但对实时视频流的时候,出现了卡顿延时的效果。在一顿度娘操作之后,...
Python推流本质是调用FFmpeg的推流进程,所以" pipe.stdin.write(img.tobytes()) "这句话报错时,可以考虑是FFmpeg没有配置的原因。
推流术语…省略。
import cv2
# subprocess 模块允许我们启动一个新进程,并连接到它们的输入/输出/错误管道,从而获取返回值。
import subprocess
# 视频读取对象
cap = cv2.VideoCapture(".../xx.mp4")
# 读取一帧
提示:文章写完后,目录可以自动生成,如何生成可参考右边的帮助文档
文章目录前言python操作ffmpeg推流0 需求:1.搭建流媒体服务2 python脚本3 验证结果:总结
提示:这里可以添加本文要记录的大概内容:
例如:随着人工智能的不断发展,机器学习这门技术也越来越重要,很多人都开启了学习机器学习,本文就介绍了机器学习的基础内容。
提示:以下是本篇文章正文内容,下面案例可供参考
python操作ffmpeg推流
0 需求:
将一个rtsp流 经过流媒体服务,推送成rtmp 供 vlc查.