Configuring and Setting up the Beacons
Test Area
Following on from our last blog, we have decided where to position our beacons in the computer lab room. We confirmed that we want to position each of the six beacons along the outside of the room (as shown in the previous blog). We have positioned the beacons in this way to ensure we can use at least 2 beacons at each reference point which will help to make our location sensing more precise. The more beacons we can get sensory information from for a single reference point, the closer we can pinpoint where the user will be.
In our simplified example, if we only used the signal strength from one beacon (Figure 1.), the user could be anywhere within the red circle. However, if we add a second and third estimote to determine where the location is, we can make our reference points more precise (Figure 2.) The middle reference point (as shown by the plus sign in the middle of the aisle) has sensory data from 3 different estimote beacons. There are only a few places that have all 3 sensor readings, which we can narrow down to the middle reference point. We chose to use machine learning instead of a reference database approach so we are not restricted to using reference points when training the model but the principal is the same. We could select the aisle we are in during training and walk around in that aisle for some amount of time and then move onto the next aisle. Overall our method incorporates some elements of fingerprinting as well as machine learning.
Following on from our last blog, we have decided where to position our beacons in the computer lab room. We confirmed that we want to position each of the six beacons along the outside of the room (as shown in the previous blog). We have positioned the beacons in this way to ensure we can use at least 2 beacons at each reference point which will help to make our location sensing more precise. The more beacons we can get sensory information from for a single reference point, the closer we can pinpoint where the user will be.
In our simplified example, if we only used the signal strength from one beacon (Figure 1.), the user could be anywhere within the red circle. However, if we add a second and third estimote to determine where the location is, we can make our reference points more precise (Figure 2.) The middle reference point (as shown by the plus sign in the middle of the aisle) has sensory data from 3 different estimote beacons. There are only a few places that have all 3 sensor readings, which we can narrow down to the middle reference point. We chose to use machine learning instead of a reference database approach so we are not restricted to using reference points when training the model but the principal is the same. We could select the aisle we are in during training and walk around in that aisle for some amount of time and then move onto the next aisle. Overall our method incorporates some elements of fingerprinting as well as machine learning.
 |
| Figure 1: 1 Estimote Beacon |
 |
| Figure 2: Using 3 Estimote Beacons |
On top of this, we decided to put the Estimotes approximately 2m off the floor. We did this to mimic the approximate height that shoppers will be holding their phone, in order to replicate how the user will be using the application to increase the accuracy of the sensor readings. We also decided to have them facing each other with the hope that they will perform better together.
To configure the beacons for this, we had to enable the iBeacon feature which basically turns the beacon into an IR blaster. The variable which we believe will be the most important is something known as the ‘advertising interval’. This is how often the beacons transmit to the device, and in complement; how often the device receives signal strength readings. This is important to us due to the nature of our app. We want the shopping list of the user to update as frequently as possible to improve the precision and overall accuracy of the app. Faster updates means the app will get more readings which will be beneficial to the training of the data.
Next Step:
We are close to getting the learning and predicting feature up and running, which will allow us to start recording data and testing the Estimote beacons for their optimal settings. For example: if we set the signal strength to a lower frequency, this may be better for the device as it will be easier to learn on highly varied signal strengths. So even though we have low signal strength for all beacons, the closer beacon will still be giving us good readings, and the further away beacons will give us extreme readings. This was an issue we noticed when watching the signal strengths for all 6 beacons, currently we have the beacons set to broadcast ~50m. The problem was that the beacon we were closest to was giving a good reading, except the other 5 readings looked extremely similar, so maybe having a closer signal fall-off point will help the overall precision of the predictor.
Written by Sam, James and Celine
To configure the beacons for this, we had to enable the iBeacon feature which basically turns the beacon into an IR blaster. The variable which we believe will be the most important is something known as the ‘advertising interval’. This is how often the beacons transmit to the device, and in complement; how often the device receives signal strength readings. This is important to us due to the nature of our app. We want the shopping list of the user to update as frequently as possible to improve the precision and overall accuracy of the app. Faster updates means the app will get more readings which will be beneficial to the training of the data.
Next Step:
We are close to getting the learning and predicting feature up and running, which will allow us to start recording data and testing the Estimote beacons for their optimal settings. For example: if we set the signal strength to a lower frequency, this may be better for the device as it will be easier to learn on highly varied signal strengths. So even though we have low signal strength for all beacons, the closer beacon will still be giving us good readings, and the further away beacons will give us extreme readings. This was an issue we noticed when watching the signal strengths for all 6 beacons, currently we have the beacons set to broadcast ~50m. The problem was that the beacon we were closest to was giving a good reading, except the other 5 readings looked extremely similar, so maybe having a closer signal fall-off point will help the overall precision of the predictor.
Written by Sam, James and Celine
Comments
Post a Comment