Why not just use the difference between the accelerometer derived altitude and Barometric altitude to determine tilt?
Many have asked why can't this be a viable method for tilt detection? The Marsa series of flight computers have used very high performance accelerometers and barometric sensors since 2008 and we have looked that the feasibility of this method ever since.
This problem is discussed in the presentation below (PDF) which is available for download.
In theory this should be possible as the difference between the Barometric altitude and Acceleration integrated derived altitude can be mathematically determinate. However the analysis shows that practical considerations make this difficult. The errors of the sensors, especially the potential error of barometric altitude estimation while the rocket has high speeds is larger than accel-baro altitude difference we need to resolve.
The summary and conclusion of the study are shown below. Much more detail can be found in the paper.
An excerpted graphic from the report is shown below.
Summary and Conclusions
• Accelerometer versus Barometric altitude difference method does not have sufficient measurement capability to be a useful staging inhibition system for moderate (10 to 20 deg) off vertical flights. False positives and negatives can happen 50% of the time.
• The largest contributor to system measurement error is the systemic error of barometric altitude lag at velocity and SAM assumption error.
• This method can be used for detection of very large (>45 deg) off axis flights.