ToF Depth Perception

Time-of-flight (ToF) technology works by sending out modulated light signal (infrared, at 940nm in our case), which bounces off objects and returns to the sensor. The sensor then measures the time taken by the light to travel back and uses this to calculate the distance (depth) of the object/scene from the sensor.

Specs	Value
Depth range limits	20cm - 5m
Depth accuracy	<1% indoors, <2% outdoors
Depth precision	<0.1%
VSCEL wavelength	940nm
Output interface:	2-lane MIPI
Ideal operating range [1]	0°C - 60°C
Operating temperature [2]	-20°C to 70°C

[1]: No visible degradation of the depth quality [2]: ToF is still fully functioningNote that the temperature ranges are for the ToF sensor itself, not for the whole camera. The SoM/RVC2 itself heats up the whole camera quite a bit, and we haven’t yet measured the ambient temperature ranges for the OAK-D ToF.

ToF vs Stereo for depth perception

Environment

Active stereo cameras can work in low-light conditions but might struggle outside, as dots from dot projector are not visible in sunlight (due to overexposure/saturation)
ToF isn’t as affected by the lighting conditions; it works in low-light environments and outside (a lot of light) as well

Depth error

Stereo depth cameras usually have below 3% of depth error. Its depth error increases exponentially with distance.
ToF sensors have a depth error of <1% indoors, <2% outdoors. Its depth error doesn’t increase exponentially with distance.

Potential Stereo issues

Repetitive textures (workaround: lower confidence threshold)
Reflective surfaces / Transparent surfaces
Occlusions (workaround: Left-Right check)
Textureless surfaces (workaround: active stereo)
Disparity jumps (workaround: subpixel mode)

Potential ToF issues

Issues when using multiple ToFs (interference)
Reflective surfaces / Transparent surfaces
Multiple-path reflections (where the light bounces off multiple surfaces before returning to the sensor)
Limited resolution: 640x480, which is considered quite high among ToF sensors.