问题描述
我目前正在努力加快车辆检测流程。我想创建一个消费者生产者模型来读取帧并处理它们。这是我试图完成的基本结构：
(Read frames from input - live stream/video) --> [Store in input buffer] -->(do vehicle detection) --> [store in vehicledetection buffer]。
鉴于从摄像头/视频读取帧是 I/O 绑定任务，因此我使用 threading.Thread，而车辆检测是 CPU 绑定进程，我为此使用 multiprocessing.Process。
import os
import cv2
import argparse
from time import sleep
import multiprocessing as mp
import threading
from queue import Queue
inputQueue = Queue(10)
class cvSkeleton():
def __init__(self,args):
self.args = args
self.frameId = 0
self.stopped=False
self.grabbed = True
s#elf.inputQueue = Queue(10)
self.vehicleDetectionQueue = mp.Queue(10)
self.outputQueue = mp.Queue(10)
self.videoBuffer = mp.Queue(10)
self.name = "output window"
if(self.args.video):
self.videoCaptureObject = cv2.VideoCapture(args.video)
sleep(2)
else:
raise Exception("No video loaded. Bad path!")
(self.grabbed,self.frame) = self.videoCaptureObject.read()
global inputQueue
def readFrame(self):
while True:
grabbed,frame = self.videoCaptureObject.read()
inputQueue.put((frame,self.frameId))
self.frameId+=1
print(f'input queue size = {inputQueue.qsize()}\n')
if(inputQueue.full()):
break
def vehicleDetection(self):
(frame,frameId) = inputQueue.get()
print(f'got frame with id {frameId}\n')
return
def numberPlateOCRdetection():
pass
# performs number plate extraction and ocr.
def displayFrame():
pass
#displays frame
def saveVideo():
pass
#takes a list of frames with frame id and merges them into a video.
if __name__ == '__main__':
import cProfile
app_profiler = cProfile.Profile()
parser = argparse.ArgumentParser(description='BitSilica Traffic Analysis Solution')
parser.add_argument('--image',help=' Full Path to image file.')
parser.add_argument('--video',help='Full Path to video file.')
parser.add_argument('--realtime',help='Camera Connected Input')
args = parser.parse_args()
#enable profiler here.
app_profiler.enable()
app = cvSkeleton(args)
readFrameProcessThread = threading.Thread(target=app.readFrame)
vehicleDetectionProcess1 = mp.Process(target=app.vehicleDetection)
readFrameProcessThread.start()
sleep(1)
vehicleDetectionProcess1.start()
readFrameProcessThread.join()
vehicleDetectionProcess1.join()
#disable profiler here.
app_profiler.disable()
profile_name = str('temp.prof'.format(os.path.basename(args.video)[0:-4]))
print("------------------------\nEnd of execution,dumping profile stats\n-------------------------")
app_profiler.dump_stats(profile_name)
错误：
input queue size = 1
input queue size = 2
input queue size = 3
input queue size = 4
input queue size = 5
input queue size = 6
input queue size = 7
input queue size = 8
input queue size = 9
input queue size = 10
^CTraceback (most recent call last):
File "skeleton.py",line 95,in <module>
Process Process-1:
readFrameProcessThread.join()
File "/home/atmadeep/anaconda3/envs/work-env/lib/python3.7/multiprocessing/process.py",line 140,in join
res = self._popen.wait(timeout)
File "/home/atmadeep/anaconda3/envs/work-env/lib/python3.7/multiprocessing/popen_fork.py",line 48,in wait
return self.poll(os.WNOHANG if timeout == 0.0 else 0)
File "/home/atmadeep/anaconda3/envs/work-env/lib/python3.7/multiprocessing/popen_fork.py",line 28,in poll
pid,sts = os.waitpid(self.pid,flag)
KeyboardInterrupt
Traceback (most recent call last):
File "/home/atmadeep/anaconda3/envs/work-env/lib/python3.7/multiprocessing/process.py",line 297,in _bootstrap
self.run()
File "/home/atmadeep/anaconda3/envs/work-env/lib/python3.7/multiprocessing/process.py",line 99,in run
self._target(*self._args,**self._kwargs)
File "skeleton.py",line 44,in vehicleDetection
(frame,frameId) = inputQueue.get()
File "/home/atmadeep/anaconda3/envs/work-env/lib/python3.7/queue.py",line 170,in get
self.not_empty.wait()
File "/home/atmadeep/anaconda3/envs/work-env/lib/python3.7/threading.py",line 296,in wait
waiter.acquire()
KeyboardInterrupt
(work-env) atmadeep:trafficAI-master/ (multiprocessing✗) $ python3 skeleton.py --video short_video.mp4 [16:50:55]
input queue size = 1
input queue size = 2
input queue size = 3
input queue size = 4
input queue size = 5
input queue size = 6
input queue size = 7
input queue size = 8
input queue size = 9
input queue size = 10
got frame with id 0
------------------------
End of execution,dumping profile stats
-------------------------
readFrame 将填充 inputQueue 到给定的 maxsize (=10) 并且 vehicleDetection 将仅读取第一帧并退出。需要帮助解决这个问题。我可能没有完全编写代码。我确实阅读了 Adrian rosebrock 的教程和许多其他人的代码，但他们都在 main 函数中使用了一个循环，这不是我想要做的，因为这样循环中的每次迭代都会产生进程并且内存会溢出。
tl：博士；在 readFrame 和 vehicleDetection 函数之间创建生产者-消费者关系。 readFrame 是一个线程，vehicleDetection 是一个进程。
解决方法
暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！
如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。
小编邮箱:dio#foxmail.com (将#修改为@）