results
Check our the Art Visualizer in action in the video below!
(Color spectrum on right- purple represents lowest frequency, right-most color represents highest frequency.)
Note that the animation slows down significantly as it goes on. This is due to Matlab being unable to keep up with the 100Hz sampling rate of the data we collected. When we measured the run time of the Matlab animation, we found that the first time step is about 0.02s, and the following steps take longer and longer, as Matlab has more data to plot with the 'hold on' command.
Contributors to Error
1st
2nd
3rd
We didn’t take into account the accumulation of errors over time in both the accelerometer and orientation data. According to this article, an angle error of even one degree will push the estimated velocity off by 1.7 m/s and the position off by 17.1 meters after 10 seconds.
Phone sensors cannot detect orientation and acceleration perfectly. For example, the gravity vector was so unstable that it was difficult to find a uniform, average gravity vector that could work for the full-length data set.
The Matlab app uses magnetic fields to obtain orientation data, so the phone's orientation sensing can be arbitrary around motors and electrical grids.
next steps
Our next steps would be to find ways to make our data collecting more accurate. The best solution would be to find better sensors that are designed for position or velocity tracking because acceleration data as a whole is made less reliable by the fact that it has to be integrated twice. However, since one of the pros of our app is that it can be made accessible to the general public by phone accelerometer and gyroscope functionality. Working with what we have, we could experiment with data recording in different locations, to avoid magnetic field interruptions. We could also make our data more accurate by accounting for the accumulation of errors by creating calibration curves that can account for the sensors' drift.
In order to make the Art Visualizer accessible to the general public, we would want to create an app for it, available on the Android and Apple App stores, where people could record data through the app, and select how they want that data to be visualized, with different color schemes, frequency intervals, and animation speeds.
If we wanted to diverge from accessibility and move towards more specific user groups, we might want to transition to using smaller sensors that can be attached more securely to a moving person or object. One of the biggest struggles we encountered while collecting data was that phones are slippery, bulky, and difficult to attach to things.