Stereo-Depth Modes

1. Active Stereo

   If there’s no texture in the images, the usual block-matching process for left-right correspondence does not work. To solve this issue, OAK Pro cameras have IR Laser Dot Projector which projects little dots on the scene and helps with stereo-matching algorithm.

   with dai.Device(pipeline) as device:
   
     #############################
     device.setIrLaserDotProjectorIntensity(0.5) # in %, from 0 to 1
     #############################
   
     # Continue...
   

   

2. Extended Disparity - A solution for short-range stereo-depth

   While matching left-right correspondance, the algorithm does not search through the whole line but just a section of the line since its computationally expensive. But, by increasing this disparity search width helps get depths of nearer objects too by adding some latency to our pipeline.

   \[Z = \frac{f.B}{d}\]

   where,

   • Z: depth

   • f: focal length

   • B: baseline

   • d: disparity

   It can be seen that depth is inversely proportional to disparity. So, if disparity increases, then depth decreases.

   pipeline = dai.Pipeline()
   stereo = pipeline.create(dai.node.StereoDepth)
   
   # Closer-in minimum depth, disparity range is doubled:
   stereo.setExtendedDisparity(True)
   

3. Subpixel Disparity

   Disparity not just discrete but can be in fraction of a pixel too, which improves precision for long-range measurements.

   pipeline = dai.Pipeline()
   stereo = pipeline.create(dai.node.StereoDepth)
   
   # Better accuracy for longer distance, fractional disparity 32-levels:
   stereo.setSubpixel(True)
   

4. Left-Right Check

   Removes incorrectly calculated disparity pixels due to occlusions at object borders (Forward Mapping + Backward Check).

   pipeline = dai.Pipeline()
   stereo = pipeline.create(dai.node.StereoDepth)
   
   # Better handling for occlusions:
   stereo.setLeftRightCheck(True)
   

References

1. Stereo Depth Node Docs

2. Configuring Stereo Depth Docs