opencv read error:[h264 @ 0x8f915e0] error while decoding MB 53 20, bytestream -7

Question

My configuration:

  ubuntu 16.04
  opencv 3.3.1
  gcc version 5.4.0 20160609
  ffmpeg version 3.4.2-1~16.04.york0

and I built opencv with:

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D PYTHON_EXECUTABLE=$(which python) -D OPENCV_EXTRA_MODULES_PATH=/home/xxx/opencv_contrib/modules -D WITH_QT=ON -D WITH_OPENGL=ON -D WITH_IPP=ON -D WITH_OPENNI2=ON -D WITH_V4L=ON -D WITH_FFMPEG=ON -D WITH_GSTREAMER=OFF -D WITH_OPENMP=ON -D WITH_VTK=ON -D BUILD_opencv_java=OFF -D BUILD_opencv_python3=OFF -D WITH_CUDA=ON -D ENABLE_FAST_MATH=1 -D WITH_NVCUVID=ON -D CUDA_FAST_MATH=ON -D BUILD_opencv_cnn_3dobj=OFF -D FORCE_VTK=ON  -D WITH_CUBLAS=ON -D CUDA_NVCC_FLAGS="-D_FORCE_INLINES" -D WITH_GDAL=ON -D WITH_XINE=ON -D BUILD_EXAMPLES=OFF -D BUILD_DOCS=ON -D BUILD_PERF_TESTS=OFF -D BUILD_TESTS=OFF  -D BUILD_opencv_dnn=OFF -D BUILD_PROTOBUF=OFF -D opencv_dnn_BUILD_TORCH_IMPORTER=OFF -D opencv_dnn_PERF_CAFFE=OFF -D opencv_dnn_PERF_CLCAFFE=OFF -DBUILD_opencv_dnn_modern=OFF -D CUDA_ARCH_BIN=6.1 ..

and use these python code to read and show:

import cv2
from com.xxx.cv.core.Image import Image

capture=cv2.VideoCapture("rtsp://192.168.10.184:554/mpeg4?username=xxx&password=yyy")
while True:
    grabbed,content=capture.read()
    if grabbed:
        Image(content).show()
        doSomething()
    else:
        print "nothing grabbed.."

Everytime, after reading about 50 frames,it will give an error like:

[h264 @ 0x8f915e0] error while decoding MB 53 20, bytestream -7

then nothing can be grabbed further,and the strange thing is:

1,comment doSomething() or
2,keep doSomething() and recording the stream from same IPCamera,then run
  code against recorded video

both cases,code works fine,can anyone tell how to solve this problem?Thank in advance!

Answer 1

I was facing this problem and I have just solved it now.

(Note: I am using python3)

I have suspected that it has something to do with the timing, because that error arises when doing expensive operations between consecutive capture.read(). Your question made me sure about this.

Also, the problem arises when I stream from my network camera (which uses H264 encoding) and there was no problem when using my laptop camera.

So the solution that worked for me is using multi-threading, with the python "threading" module. One thread streams and the other processes, while managing thread locks correctly so no read/write conflicts occur.