Wednesday, 30. September 2015

Till now we don't have a very clear idea about how to start. But the first thing we have to solve is the track recognition.

We receive the data from the 'Gyro Z' sensor. The 'basic' idea is: With the data we can build a function graph. In order to find out how the track looks like, we need to find the period of the graph. If the signal repeats (period), we know that we passed one round. (it should be possible to recognize this after 2-3 rounds)

But how to implement something? Bhuuuu... Sounds like a lot of math. Somehow we have to record all the data and then find the point where the graph repeats. (we also have to handle the noise on the signal somehow)

My first thought: Are there frameworks or libraries which can help us? (others may had similar problems to solve such things in java)

First checkout the tipps from Mr. Giersche and make some research:

  • average, standard deviation, ...
  • Spark
  • R studio
  • Mathlab

After a lot of googling and studying frameworks like spark I have somehow lost count how to solve that problem :-/ The frameworks are complex and without the knowledge of the associated math I think it's not realistic to use them meaningful.

Maybe we need an easier (naiver) approach which does not need so much math. It's 01:30 in the morning, I need a break... :-)

Update 01.10.2015 Today, Me and Roberto, we had a little talk to a math professor after lunch. In summary, he has a lot of idea's how to solve to problem, but I only understood a fraction of what he told us :D But some tipps maybe will be useful:

  • maybe use 'moving averages' to cope with the noise
  • another idea to find the track: regularly split the track into two parts and compare them. If they are equal it's maybe a candidate for a match.
  • or maybe we should use a very simple tactic first: don't use the signal and just find out if we drive straight, left or right. This gives a pattern like 'left-straight-left-left-right...' and then we can search for a repetition in that pattern. That's maybe a lot easier...
  • An open question: Are there just simple 90-degres-curves or are there different types of curves? --> ask Mr. Giersche