The inputs are sources of visual information. BitRefine Heads platform allows working with various input types. This is the description of the platform's standard inputs.

IP cameras (ONVIF)

IP cameras ONVIF source

RTSP

The RTSP input type allows reading video from the RTP stream from IP-cameras. More about streaming protocol RTSP.

RTSP input

  • frame_buffer: How many frames should pipeline keep in its internal buffer. This parameter is important for smooth display ONLY.

  • frequency: How often pipeline takes new frame from this source. (frames per second)

  • model_frequency: How often the core neural network takes frames from this particular pipeline for deep recognition. Sometimes you don't need to do object recognition in every frame. You can skip some intermediate frames and let the tracking processor bridge the gaps. Reducing model frequency can save much compute power and allow to process much more cameras on the same hardware. Value is given in seconds. (25fps is 0.04 sec)

  • url: This is the camera's RTSP URL. the common format is rtsp://login:password@ip-address:port/ You can obtain RTSP URL from camera manufacturer or from ONVIF Device manager.

  • use_rtsp_buffer: There's a buffer, provided by an RTSP device. The pipeline can take frames from the buffer or skip the buffer and read the most recent frames. Using the buffer is a preferred option if you read at 25fps, the same frame-rate as the camera streams. If you want to read at low fps (parameter frequency), the RTSP buffer should be turned off to avoid high latency.

GenICam

This input type allows connecting machine vision cameras via Ethernet. More about technology GenICam

  • frame_buffer: same as in RTSP Input
  • frequency: same as in RTSP Input
  • log_performance: same as in RTSP Input

  • model_frequency: same as in RTSP Input

  • pixel_format: Machine vision cameras allow a number of pixel formats. You can configure this parameter from here or from special camera configuration application, provided by the camera manufacturer.

  • url: This is IP-address of the GenICam device. Important to remember that GenICam relies on UDP broadcast, thus needs to be in the same subnet with the 'Head'.

Video

This input allows reading standard video files from a local or a shared folder.

Video file input

  • frame_buffer: same as in RTSP Input
  • frequency: same as in RTSP Input
  • log_performance: same as in RTSP Input

  • model_frequency: same as in RTSP Input

  • loop: After video reaches the end it will start again from the beginning. This is useful when configuring the pipeline using a test footage. If turned off, the pipeline will be stopped automatically upon reaching the end of the video file. A notification will be sent to the user.

  • video_path: This is the path to a video file. The Head's source video folder is heads_shared/src_video/.

Images

This input type allows processing batches of individual images.

Images input

  • frequency: same as in RTSP Input
  • log_performance: same as in RTSP Input
  • loop: After reaching the end of the batch pipeline will start again from the beginning.
  • path: This is the path to image files. The Head's source images folder is heads_shared/src_images/.

Dev

This input allows reading video stream from local devices connected to the Head machine, such as USB- or built-in cameras.

USB camera input

  • device_id: This is an ID of local camera. By default, starts from 0.

  • frame_buffer: same as in RTSP Input

  • frequency: same as in RTSP Input
  • log_performance: same as in RTSP Input
  • loop: After reaching the end of the batch pipeline will start again from the beginning.