Replaying

OakCamera allows users to easily use depthai-recording instead of the live camera feed to run their pipelines. This feature will send recorded frames to the OAK device. This is useful especially during development, so we can record a complex scene only once and replay it many times to fine-tune the pipeline or AI models.

Because Recording saves calibration data and can save synchronized left+right mono streams so we can achieve full depth reconstruction.

from depthai_sdk import OakCamera

with OakCamera(recording='[PATH/URL/NAME]') as oak:
    # Created CameraComponent/StereoComponent will use streams from the recording
    camera = oak.create_camera('color')

Replaying support

Replaying feature is quite extensible, and supports a variety of different inputs:

1. Single image.

2. Folder with images. Images are getting rotated every 3 seconds. Example here.

3. URL to a video/image.

4. URL to a YouTube video.

5. Path to depthai-recording.

6. A name of a public depthai-recording.

Replaying a depthai-recording

When constructing the OakCamera object we can easily replay an existing depthai-recording, which results in using XLinkIn nodes instead of ColorCamera / MonoCamera nodes.

Script below will also do depth reconstruction and will display 3D detections coordinates (XYZ) to the frame.

  from depthai_sdk import OakCamera

- with OakCamera() as oak:
+ with OakCamera(replay='path/to/folders') as oak:
      color = oak.create_camera('color')
      nn = oak.create_nn('mobilenet-ssd', color, spatial=True)
      oak.visualize(nn.out.main, fps=True)
      oak.start(blocking=True)

Live view pipeline uses live camera feeds (MonoCamera, ColorCamera) whereas Replaying pipeline uses XLinkIn nodes to which we send recorded frames.

Public depthai-recordings

We host several depthai-recordings on our servers that you can easily use in your application, e.g., OakCamera(recording='cars-california-01'). Recording will get downloaded & cached on the computer for future use.

The following table lists all available recordings:

Name	Files	Size	Notice
cars-california-01	color.mp4	21.1 MB	Source video, useful for car detection / license plate recognition
cars-california-02	color.mp4	27.5 MB	Source video, useful for car detection / license plate recognition
cars-california-03	color.mp4	19 MB	Source video, useful for license plate recognition and bicylist detection
cars-tracking-above-01	color.mp4	30.8 MB	Source video, useful for car tracking/counting
depth-people-counting-01	left.mp4, right.mp4, calib.json	5.8 MB	Used by depth-people-counting demo
people-construction-vest-01	color.mp4	5.2 MB	Used by ObjectTracker example and pedestrian reidentification demo
people-images-01	5x jpg images	2 MB	Used by people-counting demo
people-tracking-above-01	color.mp4	3.2 MB	Fisheye top-down view, useful for people tracking/counting. Fast forward/downscaled
people-tracking-above-02	color.mp4	86.4 MB	Fisheye top-down view, useful for people tracking/counting
people-tracking-above-03	color.mp4	16.7 MB	Top-down view, used by people-tracker demo
people-tracking-above-04	color.mp4	5.3 MB	Top-down view at an angle, source video here
people-tracking-above-05	CAM_A.mp4, CAM_A.mp4, calib.json	12 MB (35sec)	Top-down view, left+right stereo cameras, demo usage at replay.py

Got questions?

Head over to Discussion Forum for technical support or any other questions you might have.