Last Wednesday I went out to the field for some flight tests and gains trimming. Other people where flying their model planes there too. We usually follow a counter-clockwise pattern while flying to avoid collisions. That day I was flying especially slow, so when I took the turn, I did so on the spot and ended on the wrong lane. By the time I realized that, someone rammed me out of the air.
If I remember the even correctly, and the flight replay (LOG00212.kmz), the collision should be head-on. However, the reason I fell, was because the other plane's propeller clipped my left aileron. I managed to crash-land while doing semi-controlled spirals, losing a wing, the propeller and a wheel.
Afterwards, I tried to do some black-box analysis, with interesting results:
Firstly, here is a graph of the altitude, and the magnitude of the acceleration vector and angular speed vector:
The collision occurs at about the 120th second. I estimate the moment at 120.5s. Strangely, no acceleration or angular motion peak is noticed at that moment. Only after I try to pull up at 124s is acceleration recorded, and, of course, at the crash. The IMU is supposed to be set at +-8g limit.
The roll, pitch and yaw plots are a bit more indicative of the situation:
In the last two graphs, a violent change in direction is noticed, in contrast with the first plot.
To sum up, I didn't find the accelerometers a good source of information, in order to detect a mid-air collision. Instead, angular velocities are more reliable, with proper thresholding.