Time-of-flight cameras are often used in safety crirical applications, e.g. in anti-collision sensors for robots. It's evident, that the sensor is working correctly or, in case of a malfunction, the control system of the of the robot detects a malfunction. A serious fault in a TOF camera is the failure of one or more pixels of the TOF imager. Whereas a «stuck--at» failure is relatively easy to detect, a floating signal which can randomly take any state is not.
A pixel in an imager can be faulty in a way that it reports any level in grayscale from fully dark to fully bright. This can also be the case in a TOF imager. Thus, in a safety critical application, the distance to an object reported by a pixel is assumed to be wrong. The pixel can report the correct distance within a given tolerance band or any other distance which is not correct. Such behavior is fatal in an anti-collision sensor based on a 3D camera. The question now is, how to detect incorrect distance reporting pixels.
There are several ways to do so:
- Comparison: Comparison of the reported distance with a known distance (comparison). This can be applied e.g. in a door sensor where the sensor looks from top of the door down to the floor.
- Offset: Adding a delay into the illumination path (or the demodulation path) to impose a virtual distance shift. By subtracting the distance shift imposed by the delay, the same or a similar distance as the one without delay should be resulting.
- Scaling: Changing the modulation frequency but not changing the distance calculation parameters accordingly. This is similar like 3., but the distance shift is not fix, it is dependent on the distance value.
- Pattern: By changing the modulation or demodulation pattern, good pixels report the same (correct) distance even in a different phase sequence.
- Fill & Spill: Inject a defined amount of charge into a pixel and check the response of the pixel.
No comments:
Post a Comment
All comments are moderated to avoid spam and personal attacks.