Week 5 Journal

After last week, we were able to gather raw data and extract some signals but found them to be difficult to extrapolate readings from. One of the papers that talked about remote heart rate detection described an Independent Component Analysis (ICA) algorithm which could be used to separate a raw source into its individual components.

For this algorithm, the numpy library in Python was necessary, but OpenMV does not have numpy so I scavenged through GitHub to find a couple libraries - umatrix and ulinalg - which had most of the functions we needed from numpy and was compatible with OpenMV. After this, I worked on coding a covariance function which would be used in the ICA algorithm.

As we were working on coding this algorithm, I realized that one of the major roadblocks in the near future would be keeping track of the RGB channels with equivalent time spacing. Because the face detection algorithm only detects a face when it can, the time between each detected face is inconsistent. Due to this inconsistency, we do not get a RGB reading from the face at consistent times, causing a lot of plateaus when plotting our graphs (an example can be seen below in Figure 1). For this problem, I suspect we will have to use linear interpolation to be able to keep track of the RGB channels in a more equivalently spaced manner.

Figure 1: In between spikes, the signal for RGB channel plateaus, showing that there was no change between the previous reading and current reading. Some of these plateaus may cause a discrepancy when trying to extract the heart rate from this data.