While our algorithm focuses on analyzing the signals for the whole song, we analyzed a few moves to understand the individual motions better. Many of the motions are repeated in the song.
Assuming that the majority of moves are in time with the music, we predict that the most dominant frequencies are those that are multiples of the tempo of the song.
The following motions are taken from the song Take on Me by a-ha. The tempo of the song is 168.61 bpm (beats per minute), or 2.81 Hz.
The above graph shows the raw accelerometer data recorded for the three axes.
While the data wasn't collected at a particularly high frequency, the overall shapes of the data look similar to our predictions made above.
The graphs above show the magnitude of the accelerometer data in the frequency domain for the above motion.
In the x-axis, there are two major frequencies (2.8 Hz and 4.3 Hz).
In the y- and z- axes, there are three major frequencies (1.4 Hz, 2.8 Hz, 4.3 Hz).
The major peaks are multiples of the tempo. The peaks at around 2.8 Hz are for the movements that land on the beat (the downward motion), while the peaks at at 1.4 Hz correspond to the moves that occur every two beats (the sweeping from side to side). While the data doesn't match our predictions exactly, we recognize that the dominant frequencies are multiples of the tempo, suggesting that these motions were made with intent to line up with the beat of the song.
Note: the prominent spike at 0 Hz in the y-axis plot is related to the acceleration of gravity, which is constantly picked up by the accelerometer.
The above graph shows the raw accelerometer data recorded for the three axes.
While the data wasn't collected at a particularly high frequency, the overall shapes of the data look similar to our predictions made above; for example, the y-axis acceleration remains fairly constant while the z-axis acceleration emulates that sinusoidal wave.
The graphs above show the magnitude of the accelerometer data in the frequency domain for the above motion.
In the x-axis, there is a major frequency at about 0.7 Hz, as predicted.
In the y- and z- axis, there is a major frequency at 0.54 Hz which differs from our predictions.
The significant peak in the x-axis acceleration at 0.7 Hz aligns with our prediction. The frequency at which the data was recorded (10 Hz) was probably not frequent enough to get the desired results, and there is a lot of noise in the data that could be filtered. Overall, though, the data presented the general shape that we anticipated when creating our motion model.