Excuse me, could anyone tell me what is the accuracy of BME280 conversion height? Can the three-axis attitude angle and altitude be maintained when the four-axis hovering? (No GPS)
The four-axis itself has a three-axis attitude angle and height information, but the XY positioning function is missing. It should be able to maintain roll pitch yaw&height when hovering, leaving only XY drift. If you play with the optical flow algorithm, is it possible to ignore the interference of roll pitch yaw & height changes on the image? If the XY error is caused by the above-mentioned reasons, then all the reasons will be attributed to the flight control and it has not been done?
The datasheet says relative accuracy is +/-12Pa, ie +/- 1m in height, not allowing for barometric changes (of course). Short-term stability is likely to be better than this.
Yes of course if you have altitude readin you can stabilize height (but there's the drift in atmospheric pressure to consider - this is why automated landing and take off require a ground proximity sensor).
I got this. Thank you! 
And of course the effect of your propellers blowing air and the wind will affect any measured pressure .
I guess in reality you get the height you want ( visually ) and the hold on whatever reading you have from the sensor , so accuracy is not so important as resolution and stability
Careful siting of the sensor will minimize this - usually stagnation pressure due to wind is tiny unless its a storm, and you wouldn't be flying in a storm. The stagnation pressure is proportional to the square of windspeed.