Neural Network free fall detection

Hello.

This is my first post for neural networks and therefore I would ask for patience:)

The task we have to accomplish is to recognize the "fall" on the basis of data from the accelerometer. With the resultant free fall acceleration vector is 0. Jak the chart looks present in graphics

I would like to inquire about this task.

Generally, with each fall it is a little different, charts are stretched in time.

I was thinking about cutting out such 70 samples signals and learning network for output to give a "1" .Learning with the teacher. 70 entries is a lot, so it seems to me that, in practice, otherwise it's carried out?

Another aspect of such a fall is detected in real time. How to accomplish this?

Might at first to collect 70 samples of the signal and feeds to the network input, then collect 70 samples.

This, however, probably do not really work? it is possible that I will analyze the samples, which will be badly cut

Because of that I should do this in the following way?

1. loading 70 samples, analyze
2. The first sample absenting, the reading from the accelerometer and analyze?

The basis of the queue?

Thank you in advance for any suggestions, I greet :)

[Greg Heath]

Insufficient info. How many samples per second? How many samples does it take to visually recognize each of the events in the data?

[Aleksander s]

50 Hz sampling. It takes 70 -80 samples to recognize event. It does not always looks exactly the same...

[Greg Heath]

Would be helpful to use or add a sample count axis.

[Aleksander s]

What do You mean exactly ? What sample count axis is ? How should it helps?