Sensor fusion at the extreme edge of an internet of things network

We are witnessing these years an exponential increase in the number of IoT devices around the world. Consequently, the amount of data generated by sensor-equipped IoT devices is increasing a lot. Edge computing came up as a solution to offload the cloud, that couldn't keep pace anymore with this evolution. The workload has hence been shifted towards the edge of the network, closer to (but not at) the IoT devices. With this evolution in mind, we propose an entire new approach where all the computations such as sensor fusion are done on the sensor-equipped devices themselves. We believe that this approach will be mandatory in the future, once the approach of performing all the computations at edge gateways or at the cloud won't be viable anymore with the billions of new IoT devices that are bound to come. To show the feasibility of our approach, we have used it with a relevant use-case where sensor fusion is done on sonar-equipped IoT devices to perform indoor person tracking at near real-time. Indoor person tracking can be very useful for monitoring older persons in nursing homes. We have implemented a resilient scalable system, using GRiSP boards as prototype IoT devices, that allows people detection of up to 2 people using trilateration. This system possesses an anti-crosstalk synchronization feature to prevent sonar crosstalk to take place. It also offers the possibility of having a live view of the surveyed area. We have performed numerous deployments of our system under different scenarios to evaluate it. Encouraging result have been obtained with our approach, that has not constituted a limit at any time during our research. The only limit we have observed was the limited detection area of the sonars we used. The results of our use-case indicate that our approach can be very useful in the future. It could also be used as a complement to existing edge computing architectures, since it provides support for gateways.