Platform

ON THIS PAGE

  • Depth Perception
  • Passive Stereo Depth
  • Active Stereo Depth
  • Time-of-Flight Depth

Depth Perception

Depth perception is a key feature of the DepthAI platform, which supports different methods of perceiving depth:
  1. Passive Stereo Depth - Used by non-Pro version of OAK-D cameras.
  2. Active Stereo Depth - Used by Pro version of OAK-D cameras.
  3. Time-of-Flight Depth - Used by OAK-ToF.

Passive Stereo Depth

Passive stereo works similarly to human vision. Our brains estimate the depth of objects based on the difference in perception between our left and right eyes. In OAK-D cameras, a stereo camera pair (left and right monocular cameras) performs a similar function. The VPU (Visual Processing Unit) of the OAK cameras does disparity matching to estimate the depth.Disparity is the pixel distance between the same point in the left and right images of the stereo pair camera. The OAK-D camera calculates disparity for every pixel in the mono frame, assigning a disparity value with some confidence level. This process occurs inside the StereoDepth node. The depth map is calculated from the disparity map using this formula.Factors affecting passive stereo accuracy/smoothness:
  • Lighting/Texture: Stereo depth depends on feature matching, which can be challenging in low light or on featureless surfaces. Active stereo can resolve both texture and lighting issues.
  • Calibration: Factory calibration is typically optimal.
  • Postprocessing filters: More information can be found here (under Depth Filters). Additional filtering can be performed on the host side as well, e.g., WLS filter.

Active Stereo Depth

The OAK Pro cameras use conventional active stereo vision (ASV). A dot projector casts many small dots in front of the device, aiding in disparity matching, particularly on low-visual-interest surfaces.The stereo matching process is the same as in passive stereo, but the dots enhance accuracy.The image above shows passive and active stereo perception against a wall. The mono image (bottom left) displays the projected dots.

Time-of-Flight Depth

While stereo perception is cost-effective and suitable for long distances, it may lack accuracy. For high-accuracy applications, Time-of-Flight (ToF) is recommended. ToF can provide sub-centimeter depth accuracy.We have developed a ToF FFC module compatible with OAK-FFC. A standalone camera with an onboard ToF sensor is planned for the future.The gif above demonstrates a high-accuracy point cloud produced using a ToF FFC module and a color camera. The ToF resolution here is 244x172.