Artificial neural networks have been reported to exhibit, and in some cases surpass, human level performance on individual rigid tasks. However, these networks remain static entities of knowledge for those specific tasks, which can lead to catastrophic forgetting ---i.e., forgetting old tasks--- when attempting to learn new tasks. The main objective of Incremental Learning (IL) is to address this issue. In order to tackle catastrophic forgetting, various approaches have been proposed so far. We survey those approaches and organize them in nine categories: regularization-based methods, exemplar replay-based methods, variational continual learning-based methods, parameter isolation-based methods, dynamic architectures-based methods, distillation-based methods, generative methods, data-free methods and unsupervised methods. Moreover, this review distinguishes between two scenarios, Task Incremental Learning (Task-IL) and Class Incremental Learning (Class-IL), and reports on the results obtained for a number of experiments that compare the performance achieved by a selection of diverse methods for both scenarios on the datasets most used by the related research community.

The use of robots to inspect large infrastructures is now a reality. They allow thorough assessments of the condition of such structures, unaffected by fatigue or lack of concentration, minimizing operator risk and reducing costs. In addition, robotic inspection ensures complete structure coverage. An example is MUSSOL, an aerial platform designed for inspecting the metallic structures of large vessels, developed and tested in various research projects. Its control architecture suits inspection tasks well, providing autonomous missions for long or repetitive tasks through a list of sequential way-points. However, the lack of a specific trajectory can compromise inspection coverage. This paper presents a trajectory follower for our aerial platform to ensure straight-line movements to way-points. The method is easily configurable, and results from simulations and real experiments demonstrate its effectiveness, even in windy conditions.

In this work, we propose a 3D frontier-based exploration solution for aerial robots, focused on indoor and closed environments. Our approach provides fast and exhaustive frontier detection by expanding two trees, one local and other global, using a custom RRT algorithm. Furthermore, to enhance the expansion of the global tree, which is also used to compute paths through the environment, key branches of the local tree are joined to this structure. Moreover, in contrast to other solutions that make use of separate local and global planning processes, our strategy relies on a single-plan strategy, which permits involving frontiers found at both levels by the decision making process. Finally, to demonstrate the effectiveness of our solution, we compare its performance with GBPlanner2 in a challenging environment by using a simulated aerial robot.

Response-threshold methods stand out among the different developed swarm-like methodologies that address the task allocation problem, which must be faced in multi-robot systems in order to assign to each robot the best task to perform at each instant of time. In many real missions the tasks have associated deadlines. However, the literature only contains a few swarm methodologies, and thus response-threshold methods, tackling tasks with deadlines. Motivated by this fact, in this paper, we propose a new task allocation strategy inspired by response-threshold methods which deals with tasks with time deadlines, models stimuli using fuzzy sets and, in addition, in which each robot makes the decision about the best task to perform through the celebrated Bellman-Zadeh fuzzy optimization technique. An extensive number of simulations have been carried out in order to evaluate the quantitative performance of the swarm system based on the new approach. The results confirm that the proposed mathematical approach is able to model the evolution of the system when tasks with deadlines are under consideration. We have also observed competitive performance on a fleet of real robots, which corroborates the results derived from the simulations.

SLAM approaches rely on loop closure strategies to correct the inconsistencies of the generated map. These inconsistencies are mainly caused by the effect of sensor noise in odometry sources. For the case of visual SLAM, loop detection typically rely on the repetitive detection and matching of texture-based keypoints. Weakly textured environments, however, can lead to scenes lacking these kind of points and, hence, poor-performing loop detectors. An alternative for these environments is the use of geometrical cues such as line segments, which are frequently present within human-made, structured environments. Under this context, in this work, we introduce a novel appearance-based loop closure detection method that integrates lines and points to enhance performance in these scenarios. For this purpose, we build an incremental Bag-of-Binary-Words scheme for each visual cue to retrieve previously seen images from the two complementary perspectives. Furthermore, we rely on a late fusion strategy to combine the image candidates resulting for both visual vocabularies. An effective mechanism to group similar images close in time is applied next to reduce the effort of the image candidate search. Finally, we propose a novel scheme to validate geometrically the loop candidates, integrating lines into the procedure. The proposed approach compares favourably with other state-of-the-art methods for several datasets.

The periodic inspection of vessels is a fundamental task to ensure their integrity and avoid maritime accidents. Currently, these inspections represent a high cost for the ship owner, in addition to the danger that this kind of hostile environment entails for the surveyors. In these situations, robotic platforms turn out to be useful not only for safety reasons, but also to reduce vessel downtimes and simplify the inspection procedures. Under this context, in this paper we report on the evaluation of a new control architecture devised to drive an aerial platform during these inspection procedures. The control architecture, based on an extensive use of behaviour-based high-level control, implements visual inspection-oriented functionalities, while releases the operator from the complexities of inspection flights and ensures the integrity of the platform. Apart from the control software, the full system comprises a multi-rotor platform equipped with a suitable set of sensors to permit teleporting the surveyor to the areas that need inspection. The paper provides an extensive set of testing results in different scenarios, under different operational conditions and over real vessels, in order to demonstrate the suitability of the platform for this kind of tasks.

Process automation is enabling a level of accuracy and productivity that goes beyond human ability, and one critical area where automation is making a big difference is quality control. In this paper, we describe a semantic segmentation solution aiming at detecting the presence of quality control elements in surgery toolboxes prepared by the sterilization unit of a hospital. In order to reduce the time required to prepare pixel-level ground truth, this work focuses on the use of weakly-supervised annotations (scribbles). Moreover, our solution integrates a clustering approach into a semantic segmentation network, thereby reducing the negative effects caused by weakly-supervised annotations. The paper describes the design process and reports on the results obtained.

Vessels are nowadays widely used for goods transportation. To ensure their physical integrity, they must be periodically surveyed, searching for defects, e.g. cracks and corrosion. The inspection process is carried out at a great cost due to the arrangements required to allow the surveyors to reach any point of the structure, while guaranteeing their safety. In recent years, the idea of introducing robots in the inspection of ships is being investigated. In this line, we propose two contributions for the estimation of 3D position inside vessels, making use of UWB technology. Since this technology is typically affected by the presence of metallic elements, we firstly propose a filtering method to improve the tag-anchor distance estimates. The second contribution consists in a novel method based on ICP to estimate the position of the UWB tag making use of point-to-sphere correspondence. Both contributions have been evaluated using several datasets taken inside a real vessel. The results show that our methods are able to provide satisfactory position estimates inside that environment.

Camera trajectory estimation has generated a lot of interest during the last decades, especially for robotic positioning. As it is well-known, outdoors positioning mainly relies on GPS, whereas visual odometry is typically preferred indoors. Nevertheless, when the environment is weakly textured, visual odometers typically degrade in performance. Facing this situation, this paper develops and tests a novel visual odometer that combines image edges and planar information to estimate the trajectory of an RGB-D camera in environments that lack texture. We also present a plane matching method based on a graph matching technique. To conclude, a comparison involving the proposed odometer for two well-known datasets and other visual odometers and SLAM systems is reported. This comparison shows that our method outperforms the others for indoor poorly-textured environments, while similar performance is obtained for indoor highly-textured environments.

Recognizing whether the current place has been visited before or not is an important task in robotic navigation, since it helps to reduce the inconsistencies produced by the inherent noise of navigation sensors. When a camera is used as input for navigation, this process is known as visual loop closure detection. Under this context, in this work we propose a loop closure detection method based on superpixels and a Bag of Words scheme. A novel image description method for superpixels is proposed. Our approach also makes use of the concept of dynamic islands, which allows us to group images close in time and to avoid images taken from the same place to compete among them as loop closure candidates. The proposed method is validated using several outdoor public image sequences and compared to other state-of-the-art solutions.

Following the success of machine vision systems for on-line automated quality and process control, in this paper we describe an object recognition solution aiming at detecting the presence of quality control elements in surgery toolboxes prepared by the Sterilization Unit of a hospital. Our solution actually consists in a two-stage arbitrarily-oriented object detection method making use of indirect regression of oriented bounding boxes parameters. The paper describes the design process and reports on the results obtained up to date.

In this paper, we propose a novel appearance-based approach for topological mapping based on a hierarchical decomposition of the environment. In our map, images with similar visual properties are grouped together in nodes, which are represented by means of an average global descriptor and an index of binary features based on a bag-of-words online approach. Each image is represented by means of a global descriptor and a set of local features, and this information is used in a two-level loop closure approach, where first global descriptors are employed to obtain the most likely nodes of the map and then binary image features are used to retrieve the most likely images inside these nodes. This hierarchical scheme enables us to reduce the search space when recognizing places, maintaining high accuracy when creating a map. Our approach is validated using several public datasets and compared against several state-of-the-art techniques. The accuracy and the sparsity of the generated maps are also discussed.

Seagoing vessels have to undergo regular visual inspections in order to detect defects such as cracks and corrosion before they result into catastrophic consequences. These inspections are currently performed manually by ship surveyors at a great cost, so that any level of assistance during the inspection process would significatively decrease the inspection cost. In this paper, we describe a novel framework for visually detecting the aforementioned defects. This framework is generic and flexible in the sense that it can be easily configured to compute the features that perform better for the inspection at hand. In this regard, inspired by the idea of conspicuity, this work considers contrast and symmetry as features for detecting defects on vessel structures. A combination operator is additionally tested in order to merge the information provided by these features and improve the detection performance. Experimental results for the different configurations of the detection framework show better classification results than state of the art methods.

Image mosaicing has gained increasing attention in the last few years, specially for robotic mapping applications. Due to the richness of the sensor data provided, several science fields require the creation of large-area image mosaics for further analysis. In this paper, we propose a novel and generic image mosaicing approach that can produce seamless composites under different configurations in a reasonable amount of time. Our approach is based on a multi-threaded architecture which allows us to execute the different steps of the algorithm simultaneously. To find the topology of the environment, we use a visual index based on a Bag-of-Binary-Words scheme, which is built in an online manner, and thus avoids the classic training step. Our approach is validated under different environments and camera configurations, showing that it can be used on several scenarios producing coherent results in all of them. Furthermore, the implementation of the algorithm is made public to the community.

Seagoing vessels have to undergo regular visual inspections in order to detect the typical defective situations affecting metallic structures, such as cracks and corrosion. These inspections are currently performed by ship surveyors manually at a great cost. To make ship inspections safer and more cost-efficient, this paper presents a Micro-Aerial Vehicle (MAV) intended for visual inspection and based on supervised autonomy. On the one hand, the vehicle is equipped with a vision system that effectively teleports the surveyor from the base station to the areas of the hull that need inspection. On the other hand, the MAV is the result of a complete redesign of a visual inspection-oriented aerial platform that we proposed some years ago, with the aim of introducing the surveyor in the control loop and, in this way, enlarge the range of inspection operations that can robustly be carried out. Another goal is to make the platform as usable as possible for a non-expert. All this has been accomplished by means of the definition of different autonomous functions, including obstacle detection and collision prevention, and extensive use of behavior-based high-level control. The results of some experiments conducted to assess both the performance and usability of the platform are discussed at the end of the paper.

Vessel maintenance entails periodic visual inspections of internal and external parts of the hull in order to detect the typical defective situations affecting metallic structures. Nowadays, robots are becoming more and more important regarding these inspection tasks, since they can collect the requested information and, thus, prevent humans from performing tedious, and even dangerous tasks because of places hard to reach for humans. A Micro Aerial Vehicle (MAV) fitted with vision cameras can be used as part of an automated or semi-automated inspection strategy. The resulting collection of individual images, however, does not permit the surveyor to get a global overview of the state of the surface under inspection, apart from the fact that the defects can be separated into different consecutive images. Image mosaicing can certainly help in this case. To this end, in this paper, we propose a novel image mosaicing approach able to deal with this kind of scenarios. Our solution employs a graph-based registration method from which relevant topological relationships between (overlapping) images are found. This graph is built according to a visual index based on a Bag-of-Words (BoW) scheme making use of binary descriptors for speeding up the image description process. At the end of the paper, we report about the results of a number of experiments that validate our approach, including the outcome of defect detectors working directly over the mosaic.

The use of saliency mechanisms for defect detection is discussed in this work. We consider defects on regular surfaces as conspicuous areas that catch the attention of the surveyors. Following this approach, we propose the use of the Bayesian framework SUN, devised to provide saliency information based on natural statistics, to combine information about the visual appearance of the surface under inspection, to finally indicate where the defects (if any) are located. The visual information is suggested to be based on features commonly used to predict human eye fixations: contrast and symmetry. We demonstrate that these two features provide a description of the surface that can be used to indicate whether it is defective or not. Our approach is assessed using a publicly available image dataset containing a variety of surfaces with defective areas. The performance of the defect detector is evaluated through cross-validation and successful results are obtained.

We propose an appearance-based loop closure detection algorithm based on binary features and a Bag-of-Words scheme. Unlike other approaches that build the visual dictionary offline, we introduce an indexing method for binary features, which, in combination with an inverted index, enable us to obtain loop closure candidates in an online manner. These structures are used in a discrete Bayes filter to select final loop candidates and to ensure temporal coherency between predictions. Our approach is validated using two publicly available datasets of outdoor environments and compared with the state-of-the-art FAB-MAP algorithm, showing very promising results and demonstrating that binary features can be used for visual loop closure detection.

Vessel maintenance entails periodic visual inspections of internal and external parts of the hull in order to detect the typical defective situations affecting metallic structures, such as cracks, coating breakdown, corrosion, etc. The main goal of the EU-FP7 project MINOAS is the automation of the inspection process, currently undertaken by human surveyors, by means of a fleet of robotic agents. This paper overviews a semi-autonomous approach to the inspection problem consisting of an autonomous Micro Aerial Vehicle (MAV) to be used as part of this fleet and which is in charge of regularly supplying images that can effectively teleport the surveyor from a base station to the areas of the hull to be inspected. Specific image processing software to analyze those images and assist the surveyor during the repair/no repair decision making process is also contributed. The control software approach adopted for the MAV, including self-localization and obstacle avoidance, is described and discussed, and experimental results in this regard are as well reported.

The paper proposes a method that uses topological information to guide the path search in any 2D workspace. As a first contribution, we have extended the method proposed by Jenkins to generate the topological information by taking into consideration the constraints of the workspace during the construction of the topological environment, which is then used to compute topological paths. As a second contribution, a planner based on the Rapidly-exploring Random Tree (RRT), called Homotopic RRT (HRRT) is proposed to use the topological information to guide the path search in the workspace. Finally, simulated and real results with an Autonomous Underwater Vehicle (AUV) are presented showing the feasibility of the proposal. Comparison with well-known path planning algorithms has also been included.

This paper proposes a mobile robot localization approach based on a monocular vision system. The proposal operates in three main steps. First, image features are extracted from frames gathered at different positions. Second, the obtained features are classified as obstacles or ground points. Third, under the assumption of a flat floor, the ground point coordinates are computed and used to perform localization. The experimental results, performed both in simulation and real environments, validate the proposal.

Periodic visual inspection of the internal and external parts of vessels hull is typically performed by trained surveyors at great cost. Assisting them during the inspection process by means of mechanisms capable of defect detection would certainly decrease inspection cost. With this aim, this paper presents a corrosion detection algorithm built around a weak classifiers cascade scheme and reports on its performance. As a secondary contribution, a crack detection approach guided by the output of the corrosion detector is also proposed. As a result, false positives rate is reduced as well as the computation time.

Vessel maintenance entails periodic visual inspections of internal and external parts of the vessel hull in order to detect the typical defective situations affecting metallic structures, such as cracks, corrosion, etc. This paper presents the first steps towards a roboticized visual inspection system aiming at automating as much as possible the whole inspection process currently undertaken by human surveyors. Occurrence of defects are first outlined, followed by a brief description of the whole system. Next, details for one of the robotic platforms to be used are given, as well as for the image processing algorithms dealing with the visual detection of defects.

Reactive navigation strategies for mobile robots base their success on the ability to discriminate obstacles from the ground in the vicinity of the robot and many of these strategies are based uniquely on the computation of quantitative information. In this paper we describe a new method to build qualitative local occupancy grids that are used in a new vision-based navigation strategy addressed to mobile robots to explore safely unknown environments. The process includes a new feature classifier based on the inverse perspective transformation to discriminate object from ground points and a method to determine angle and range with respect to the camera in the world coordinate system.

In recent years, anytime algorithms have shown to be a good solution for planning a path in domains with severe restrictions regarding the time for deliberation. They typically operate by quickly finding a highly suboptimal path first, and then improving it until the available time runs out. In this paper, we propose a novel anytime approach called ABUG that performs much more efficiently than the competing strategies. ABUG is based on an improved version of a member of the popular family of algorithms known as Bug. A formal analysis of the planner is provided and several relevant properties of ABUG are identified. Besides, as done in some heuristic-based anytime approaches, we define bounds on the quality/length of the paths returned by the algorithm. Finally, in order to demonstrate the computational savings associated with the proposal, a comparative study involving a set of well-known path-planning techniques is also carried out.

Reactive mobile robot navigation based on potential field methods has shown to be a good solution for dealing with both unknown and dynamic environments, where timely responses are required. Unfortunately, the complexity of the tasks which can be successfully carried out is restricted by the inherent shortcomings of the approach such as trapping situations due to local minima, difficulties passing among closely spaced obstacles, oscillations in narrow corridors, etc. The first of the aforementioned limitations was fully overcome in J. Antich and A. Ortiz (2005) by applying the two new principles of traversability and tenacity in the context of artificial potential fields. As a result, navigation was achieved in very difficult scenarios, even including maze-like environments. In this paper, continuing that previous work, it is intended to step forward by ensuring, whenever possible, that the target point is attained for any mission, setting, at the same time, an upper bound regarding the length of the resultant robot's path

Reactive mobile robot navigation based on potential field methods has shown to be a good solution for dealing with unknown and dynamic environments, where timely responses are required. Unfortunately, the complexity of the tasks which can successfully be carried out is restricted by the inherent shortcomings of the approach such as trapping situations due to local minima, difficulties passing among closely spaced obstacles, oscillations in narrow corridors, etc... This paper proposes a novel strategy which overcomes totally the first limitation and partially the others by computing an adaptive navigation function on the basis of such artificial potential fields. As a result, navigation is achieved in very difficult scenarios such as maze-like environments. A comparative study on the path length performance of our proposal with regard to other algorithms from the related literature is also presented. Both simulation and real tests are performed.

Nowadays, the surveillance and inspection of underwater installations, such as power or telecommunication cables and pipelines, are carried out by trained operators who, from the surface, control a Remotely Operated Vehicle (ROV) with cameras mounted over it. This is a tedious, time-consuming, and expensive task, prone to errors mainly because of loss of attention or fatigue of the operator and also due to the typical low quality of seabed images. In this article, a reactive control architecture for visually guiding an Autonomous Underwater Vehicle (AUV) to detect and track a cable, or pipeline, laid on the seabed is presented together with a set of experimental results. Such results were obtained by using a three-dimensional simulation environment that incorporates the hydrodynamic model of a real underwater vehicle called GARBI. Additionally, a rapidly convergent off-line tuning method based on evolutionary theory is proposed and discussed. © 2005 Wiley Periodicals, Inc. Int J Int Syst 20: 477–498, 2005.

The irradiance measurement performed by vision cameras is not noise-free due to both processing errors during CCD fabrication and the behaviour of the electronic device itself. A proper characterization of sensor performance, however, allows either removing the resulting noise from the image or accounting for it within image processing algorithms. This paper proposes new methods for estimating the distribution parameters of two of such sources of noise: dark current and the so-called fixed pattern noise. Since both methods require knowledge about the scene illumination, an estimation method using a calibrating sphere is also presented. This method models illumination as the combination of directional and ambient lighting. Experimental results can be found at the end of the paper.

Most shape from shading (SFS) algorithms assume that objects in the scene are Lambertian reflectors, i.e., they do not absorb any radiation for any wavelength. However, real objects reflect only part of the incoming radiation, so that these algorithms usually perform poorly on images with non-uniform reflectivity. In this paper an adaptation for such algorithms is proposed. The strategy mainly consists of guiding the SFS algorithm through the image in order not to violate the typical albedo uniformity assumption. To do this, the pixels in the image are iteratively grouped into regions coming from the same scene until there are no more adjacent pixels satisfying this condition. The decision about the aggregation of a new pixel to a region is based on estimating its albedo and comparing it with the albedo of the region. During this process, SFS is applied only to the considered region and its results are used in the albedo estimation procedure. Some experimental results with synthetic and real images are discussed.

Nowadays, the surveillance and inspection of underwater installations, such as power and telecommunication cables and pipelines, is carried out by operators that, being on the surface, drive a Remotely Operated Vehicle (ROV) with cameras mounted over it. This is a tedious and high time-consuming task, easily prone to errors mainly because of loss of attention or fatigue of the human operator. Besides, the complexity of the task is increased by the lack of quality of typical seabed images. In this study, the development of a vision system guiding an Autonomous Underwater Vehicle (AUV) able to detect and track automatically an underwater power cable laid on the seabed has been the main concern. The vision system that is proposed tracks the cable with an average success rate above 90%. The system has been tested using sequences coming from a video tape obtained in several tracking sessions of various real cables with a ROV driven from the surface.