Video-sensing characterization for hydrodynamic features: Particle tracking-based algorithm supported by a machine learning approach

The efficient and reliable monitoring of the flow of water in open channels provides useful information for preventing water slow-downs due to the deposition of materials within the bed of the channel, which might lead to critical floods. A reliable monitoring system can thus help to protect properties and, in the most critical cases, save lives. A sensing system capable of monitoring the flow conditions and the possible geo-environmental constraints within a channel can operate using still images or video imaging. The latter approach better supports the above two features, but the acquisition of still images can display a better accuracy. To increase the accuracy of the video imaging approach, we propose an improved particle tracking algorithm for flow hydrodynamics supported by a machine learning approach based on a convolutional neural network-evolutionary fuzzy integral (CNN-EFI), with a sub-comparison performed by multi-layer perceptron (MLP). Both algorithms have been applied to process the video signals captured from a CMOS camera, which monitors the water flow of a channel that collects rain water from an upstream area to discharge it into the sea. The channel plays a key role in avoiding upstream floods that might pose a serious threat to the neighboring infrastructures and population. This combined approach displays reliable results in the field of environmental and hydrodynamic safety.