ProcessedPiRecorder

A multiprocessed wrapper of PiCamera for simplified deployment of high framerate computer vision on raspberry pi.

Installation

  pip install ProcessedPiRecorder

Requires

I'm sure it would work with other versions, but these are the ones used during dev.

Basic Usage

You have to initialize the recorder and then tell it when to start recording.

Initialize:

  from ProcessedPiRecorder import ProcessedPiRecorder as ppr

  myRecorder = ppr(tif_path=None, x_resolution=0, y_resolution=0, scale_factor=1, framerate=0, 
                   rec_length=0, display_proc='camera_reader', stereo=False,
                   timestamp=False, report_Hz=False, monitor_qs= False, Hz_buffer=10,
                   callback=None, blocking=True,  tif_compression=6, 
                   latency_log=None, aquisition_log=None,  cb_log=None)

Start recording

  myRecorder.recordVid()

Queues and Callbacks

ProcessedPiRecorder works by separating the acquisition, computer vision, and file encoding tasks across multiple python processes using the standard python multiprocessing library. These processes pass frames using multiprocessing.Queue objects which are managed by QueueHandler objects so you don't muck them up.

Queue Structure

Callback structure

Computer vision can be easily added by means of a callback function. Frames are recieved from and placed into the queues using a ppr.QueueHandler object which also provides a buffer of frames for applications that require a series of frames.

   import ProcessedPiRecorder as ppr

   callback_fucntion(queue1, queue2, cb_queue, cb_logger):
        #inits    
        handler = ppr.QueueHandler(queue1, queue2, 2)

        #infinite loop
        while True:
              #get frame
              frame = handler.get()

                    #make sure the handler has a new frame and the buffer has filled
                    if not handler.empty and handler.full_buffer:

                          #Do something
                          do_something_to(frame)

                          #Save the frame
                          handler.put()

StereoPi support

The StereoPi is cool, but using standard PiCamera you can't save a highframerate video to file without dropping frames, ProcessedPiRecorder fixes that. Be aware that the scale_factor parameter must be used to down sample the frames. I use the following parameters as a starting point for high framerate acquisition (~28Hz) on stereopi:

  x_resolution=1280, y_resolution=480, scale_factor=0.3, framerate=25

Contributors

This code was written and is maintained by Matt Davenport (mdavenport@rockefeller.edu).