Recherche avancée

Sur d’autres sites (9104)

• Create 1 RTSP stream from 3 RTSP streams on Raspberry Pi 3

  29 mai 2020, par user3260912

  So this is interesting question and was wondering if anyone may have any insights to how I could do this. I currently have 6 IP cameras and run a Java process that opens ffmpeg to rip the RTSP streams from those 6 IP cameras, save a image to RAM, and then use ImageMagick to convert those files to a collage JPG image so I have all IP cameras in one image. That file then updates as rapidly as possible using space in /dev/shm - in reality, about 6 FPS. But, it uses 45-50% CPU on 6 cores.

  



  I'm looking for a way to reduce some CPU load off my main computer, though. I have 2 Raspberry Pi model 3Bs and am thinking I could maybe put them to good use. Not sure how the performance would be, but willing to test this.

  



  What I'm wanting to do is this :

  



  

  1. Use ffmpeg to pull down images from 3 of the IP camera RTSP streams on each Raspberry Pi into /dev/shm

  


  2. Using ImageMagick, montage the temp images pulled, into /dev/shm

  


  3. Create a RTSP stream on each Raspberry Pi of that montaged image in /dev/shm

  


  4. Use my desktop to pull down the RTSP of the collaged images and collage those to the same format I do today (only using 2 RTSP stream threads, instead of 6 to do this.)

  


  



  Is there a way to set ImageMagick to set image output format as mjpeg2 or have ffmpeg create a rtsp stream off the rapidly updating JPEG image file ?

  


• dnn/native : add native support for 'mul'

  11 avril 2020, par Guo, Yejun

  dnn/native : add native support for 'mul'
  it can be tested with model file generated from above python script :
  import tensorflow as tf
  
  import numpy as np
  
  import imageio
  in_img = imageio.imread('input.jpg')
  
  in_img = in_img.astype(np.float32)/255.0
  
  in_data = in_img[np.newaxis, :]
  x = tf.placeholder(tf.float32, shape=[1, None, None, 3], name='dnn_in')
  
  z1 = 0.5 + 0.3 * x
  
  z2 = z1 * 4
  
  z3 = z2 - x - 2.0
  
  y = tf.identity(z3, name='dnn_out')
  sess=tf.Session()
  
  sess.run(tf.global_variables_initializer())
  graph_def = tf.graph_util.convert_variables_to_constants(sess, sess.graph_def, ['dnn_out'])
  
  tf.train.write_graph(graph_def, '.', 'image_process.pb', as_text=False)
  print("image_process.pb generated, please use \
  
  path_to_ffmpeg/tools/python/convert.py to generate image_process.model\n")
  output = sess.run(y, feed_dict=x : in_data)
  
  imageio.imsave("out.jpg", np.squeeze(output))
  Signed-off-by : Guo, Yejun <yejun.guo@intel.com>
  

  • [DH] libavfilter/dnn/dnn_backend_native_layer_mathbinary.c
  • [DH] libavfilter/dnn/dnn_backend_native_layer_mathbinary.h
  • [DH] tools/python/convert_from_tensorflow.py
  • [DH] tools/python/convert_header.py

• Efficient way to stream a sequence of frames

  3 juin 2020, par DoriHp 0

  I'm facing with an issue : I implemented a device to detect stranger for my home, which includes inputs from some IP camera and use a tensorflow model to process frame got from them.

  



  Now I want to build a dashboard (use Flask or Django - python framework as backend) to streaming the processed frames I got from the system, and if possible, do some transform on them (such as stack multi frames into one, etc), and run the server so I can watch it from distances. Currently, I'm sending frame by frame as independent images but it costs so much bandwidth. I read how h264 encoder work and felt very exicted about it. Now, the question is, how can I use h264 or any encoder like it transfer my data and reduce the bandwidth ?