The detection of oil spills in water is a frequently researched area, but most of the research has been based on very large patches of crude oil on offshore areas. We present a novel framework for detecting oil spills inside a port environment, while using unmanned areal vehicles (UAV) and a thermal infrared (IR) camera. This framework is split into a training part and an operational part. In the training part, we present a process for automatically annotating RGB images and matching them with the IR images in order to create a dataset. The infrared imaging camera is crucial to be able to detect oil spills during nighttime. This dataset is then used to train on a convolutional neural network (CNN). Seven different CNN segmentation architectures and eight different feature extractors are tested in order to find the best suited combination for this task. In the operational part, we propose a method to have a real-time, onboard UAV oil spill detection using the pre-trained network and a low power interference device. A controlled experiment in the port of Antwerp showed that we are able to achieve an accuracy of 89% while only using the IR camera.

In order to determine camera parameters, a calibration procedure involving the camera recordings of a checkerboard is usually performed. In this paper, we propose an alternative approach that uses Gray-code patterns displayed on an LCD screen. Gray-code patterns allow us to decode 3D location information of points of the LCD screen at every pixel in the camera image. This is in contrast to checkerboard patterns where the number of corresponding locations is limited to the number of checkerboard corners. We show that, for the case of a UEye CMOS camera, the precision of focal-length estimation is 1.5 times more precise than when using a standard calibration with a checkerboard pattern.

To automatically evaluate the ergonomics of workers, 3D skeletons are needed. Most ergonomic assessment methods, like REBA, are based on the different 3D joint angles. Thanks to the huge amount of training data, 2D skeleton detectors have become very accurate. In this work, we test three methods to calculate 3D skeletons from 2D detections: using the depth from a single RealSense range camera, triangulating the joints using multiple cameras, and combining the triangulation of multiple camera pairs. We tested the methods using recordings of a person doing different assembly tasks. We compared the resulting joint angles to the ground truth of a VICON marker-based tracking system. The resulting RMS angle error for the triangulation methods is between 12° and 16°, showing that they are accurate enough to calculate a useful ergonomic score from.

We develop a new gradient-based optimization strategy for robotic inspection planning. The algorithm starts from an initial measurement path and evolves this path to decrease path length while maintaining sensor coverage. The approach is powerful enough to handle inspection tasks of complex objects under articulated manipulator constraints. We show the effectiveness of the algorithm in two complex inspection tasks.

In this work, we studied the potential of the visible, near-infrared, and shortwave infrared wavelength regions for monitoring oil spill incidents using optical reflectance. First, a simple physical model was designed for accurate oil thickness and volume estimation using optical reflectance. The developed method was made invariant to changes in acquisition and illumination conditions. In the next step, an algorithm based on an artificial neural network was designed to detect spilled oil. The training samples that are required to optimize the parameters of the network were generated by utilizing the proposed physical model. To validate the method, experiments were conducted in laboratory and outdoor scenarios for detection and thickness/volume estimation on four different oil types. In particular, we developed hyperspectral datasets of oil samples with varying thickness between 500 µm and 5000 µm acquired using two different sensors, an Agrispec spectrometer and an Imec snapscan shortwave infrared hyperspectral camera, in strictly controlled experimental settings. To demonstrate the potential of the proposed method in outdoor environments using solely the visible wavelength region, we monitored the evolution of artificially spilled oil in an outdoor scene with an RGB camera mounted on a drone.