Recherche avancée

Sur d’autres sites (3725)

• Combining JavaCV and openCV

  13 juin 2014, par mister-viper

  I have the following problem :
  I have an Android application which uses native OpenCV code. In a first step, the frames which were edited by OpenCV came from the camera. Then they were processed and drawn on the display.

  However, my requirements now have changed. The frames which have to be edited come from a video file stored on the SD card. They must be processed by the openCV code and then stored in a new video file.

  After reading some a lot of stuff, I recognized that Android has no built-in stuff for correctly reading a video file frame by frame and allowing to process the frames while doing so. On a computer OpenCV has the VideoCapture function. But this does not work on Android as openCV has no ffmpeg that comes with it.

  After reading more stuff, I found that JavaCV comes with an FFMPEGFrameGrabber and also an FFMPEGFrameRecorder. So, I implemented everything which now allows me to grab single frames from a video, obtain an IplImage frame and store this frame in a new video.

  Now the problem :
  During obtaining and storing the IplImage frame must be processed using the original OpenCV code as it is not feasible to port the complete code to JavaCV.

  So in a first place I wrote a small test JNI function which gets the address of a MAT object and draws a small circle on it.

  extern "C" {
  
  JNIEXPORT void JNICALL Java_de_vion_postprocessing_step2_EyeTracking_editFrame(
  
      JNIEnv*, jobject, jlong thiz, jlong addrRgba) {
  
  //Convert the mat addresses into the objects
  
  Mat& rgbFrame = *(Mat*) addrRgba;
  
  
  
  Point2i scaledSmoothPoint(100,100);
  
  circle(rgbFrame, scaledSmoothPoint, 20, YELLOW, -1);
  
  }

  As I read that IplImage extends CvArr I just call the function within in my code as follows :

  captured_frame = grabber.grab();
  
  if (captured_frame == null) {
  
      // no new frames
  
      break;
  
  }
  
  editFrame(captured_frame .address());

  However, I now get the following error :

  06-12 18:58:23.135: E/cv::error()(6498): OpenCV Error: Assertion failed (cn <= 4) in
  
                      void cv::scalarToRawData(const Scalar&, void*, int, int), file 
  
                      /home/reports/ci/slave_desktop/50-SDK/opencv/modules/core/src/matrix.cpp, line 845
  
  06-12 18:58:23.135: A/libc(6498): Fatal signal 6 (SIGABRT) at 0x00001962 (code=-6),
  
                      thread 6526 (AsyncTask #1)

  Finally, me question :
  How can I process the IplImage frame using nativeOpenCV and finally store this IplImage frame then in the video recorder.

  I am also open to new Ideas which do not necessarily require JavaCV as long as I do not have to write the FrameGrabber and FrameRecorder my self.

  Best regards,
  André

• avcodec/dirac_arith : Fix build with PIC and stack-check options

  9 avril 2014, par Anthoine Bourgeois

  avcodec/dirac_arith : Fix build with PIC and stack-check options
  Fixes Ticket3540
  The function dirac_get_arith_bit in libavcodec/dirac_arith.h can’t be
  
  built with PIC and check-stack because the asm code needs 6 registers
  
  and PIC and check-stack options take 1 each and x86 is quite limited
  
  in this area.
  Signed-off-by : Michael Niedermayer <michaelni@gmx.at>
  

  • [DH] libavcodec/dirac_arith.h

• Ffmpeg - multi overlay - where is the mistake ?

  22 décembre 2018, par georgmann

  Four videos should be evenly spread over the entire area.
  I am trying to use this code, and I cannot find an error.
  Size of the logo.png file 1920x1080.

  ffmpeg -loop 1 -i logo.png -i 01.mp4 -i 02.mp4 -i 03.mp4 -i 04.mp4 -filter_complex "[1:v]scale=960x540[a]; [2:v]scale=960x540[b]; [3:v]scale=960x540[c]; [4:v]scale=960x540[d]; [0:v][a]overlay=0:0:shortest=1[e]; [0:v][b]overlay=960:0:shortest=1[e]; [0:v][c]overlay=540:0:shortest=1[e]; [e][d]overlay=0:960[video]" -y -ss 00:00:00 -t 00:00:05 -b 3000k -map "[video]" 05.mp4

  Audio track from video "01.mp4"