<p>As more and more computational power becomes available at increasingly affordable prices, the last years have seen a veritable explosion in the number of sensors and interconnected devices. This evolution is well known and often referred to as the 4th industrial revolution, or the IoT. The water sector, albeit often conservative in adopting new technologies, will profit from this continued digitalisation in various ways.</p><p>In this work we focus on the vision of covering entire sewer systems by tightly knit sensor networks which can process the generated amount of data simultaneously. Given the large number of sensors required, the only possibility to implement such a network is keeping costs as low as possible for the individual devices or use already existing sensors in multiple ways (e.g., traffic cameras helping in flood detection).</p><p>Using hardware of the Raspberry Pi ecosystem, currently retailing at less than 100$, we collected continuous video footage of an artificial open channel in a laboratory setting and used a deep neural network to extract the water level and surface velocity. The measurement accuracy of the prediction algorithm was then compared to conventional flow sensors to assess the practicality of this approach. Preliminary results in a laboratory setting indicate a sufficient prediction accuracy of the water level for engineering uses but further work is needed to verify this in a long-term field study.</p><p>After this initial stage, deploying the sensor in a real-world setting as part of the B-WaterSmart project is planned. Apart from verifying the results under real conditions, we will then be able to assess the long-term behaviour of this approach. This includes an evaluation of the maintenance effort. As the sensor is not in direct contact with the sewage, the typical need for frequent cleaning should be greatly reduced, which in turn is expected to further lower the costs.</p><p>We argue that if such a cheap sensor can ultimately be established as a viable alternative to more conventional flow sensors, the vision of sewer networks covered entirely by sensors, could become more attainable in practice.</p>
Fault-Tolerant Optical Flow Sensor/SINS Integrated Navigation Scheme for MAV in a GPS-Denied Environment
