In our prior work in [11] we have identified triggering threshold (denoted by TR th) and de-triggering threshold (denoted by DTRth) for this ratio hat provide best discriminatory power as TRth = 1.75 and DTRth = 1.5 and with window sizes of STA = 0.1 second and LTA = 10 seconds for the explosion parameters.
In our experiments, the phone will only retain accelerometer readings that meet the above criteria, while deleting all other readings. Only the retained accelerometer readings are leveraged for post-processing to develop the model which will be discussed in the next step. A total of 52 individual blast event data-sets were obtained as a result of this step (following the above procedure), and the corresponding distances and intensities (ground-truth values) from the experiments were also recorded for subsequent training and testing
Step 2: Labeling: With experimental data-sets obtained, we are ready to proceed with designing the model. Clearly, the first step to do in this regard is labeling the data-sets appropriately based on ground truth. The distance and intensity are denoted as world states, which we aim to estimate.