R&D highlights edition 2019

Enabling technologies CANDID CAMERA – USING VIDEO AND ARTIFICIAL INTELLIGENCE AS MEASUREMENT INSTRUMENTS T he field of Computer Vision has advanced rapidly in recent years with the introduction of artificial intelligence and video imagery for tasks such as facial recognition and self-driving cars. At Deltares, state-of- the-art computer vision algorithms have been successfully applied in our physical model facilities, providing a non-intrusive measurement instrument. The development of these techniques at Deltares has been funded by a joint effort from Hydralab+, the CoastalFoam JIP, SO Coastal & Offshore Engineering and SO Measurement Techniques. The developed algorithm uses a convolutional neural network (CNN) to perform image segmentation. When a CNN has been trained for image segmentation, the network predicts one of the predefined classes for each pixel in the image. Typical examples of classes in this case are water, air, sand and rock armour layers. The CNN itself consists of many subsequent layers that perform filtering and compression. This results in an image with fewer pixels but far more complex information. A trained CNN can recognise features from these highly compressed complex images and match the associated classes. A CNN is a Deep Learning network that needs to be trained using examples of the desired outcome. In practice, this means that a number of images (+/- 100) need to be annotated to indicate the different classes they contain (in other words to indicate the water, sand, etc. in the image). These annotated images form the training and testing set for the CNN. The training set is used to train the CNN by iteratively optimising the weightings given to the different layers and it serves as an unseen control group to safeguard the predictive quality of the network. The technique has already been applied in both the Delta Flume and Scheldt Flume facilities. Applications include measuring wave heights, run-up, overtopping and changes in the bed level. The technique provides a non-intrusive measurement instrument that can capture hard- to-measure phenomena such as run-up velocities or morphological development during a test. The potential for additional applications includes field measurements of run-up and tracking the movement of flexible nature-based solutions. Contact: Joost den Bieman, Joost.denBieman@deltares.nl, t +31 (0)6 46911198 Menno de Ridder, Menno.deRidder@deltares.nl, t +31 (0)6 51789665 The field of Computer Vision has made major advances in recent years using artificial intelligence and video for tasks like facial recognition and self-driving cars. At Deltares, these techniques have been successfully applied in our physical model facilities to measure wave heights, run- up, overtopping and changes in bed level using video cameras. Example of a time stack with annotated run-up level (green line) Validation of wave spectrum from video (red) using regular wave height meter (black) Schematic layout of the CNN, after Badrinarayanan et al. (2017) 61 R&D Highlights 2019