An annotated water-filled, and dry potholes dataset for deep learning applications.

Potholes have long posed a challenging risk to automated systems due to their random and stochastic shapes and the reflectiveness of their surface when filled with water, whether it is "muddy" water or clear water. This has formed a significant limitation to autonomous assistive technologies such as Electric-Powered Wheelchairs (EPWs), mobility scooters, etc. due to the risk potholes pose on the user's well-being as it could cause severe falls and injuries as well as neck and back problems. Current research proved that Deep Leaning technologies are one of the most relevant solutions used to detect potholes due to the high accuracy of the detection. One of the main limitations to the datasets currently made available is the lack of photos describing water-filled, rabble-filled, and random coloured potholes. The purpose of our dataset is to provide the answer to this problem as it contains 713 high-quality photos representing 1152 manuall-annotated potholes in different shapes, locations, colours, and conditions, all of which were manually-collected via a mobile phone and within different areas in the United Kingdom along with two additional benchmarking videos recorded via a dashcam.

Optimisation of Deep Learning Small-Object Detectors with Novel Explainable Verification.

In this paper, we present a novel methodology based on machine learning for identifying the most appropriate from a set of available state-of-the-art object detectors for a given application. Our particular interest is to develop a road map for identifying verifiably optimal selections, especially for challenging applications such as detecting small objects in a mixed-size object dataset. State-of-the-art object detection systems often find the localisation of small-size objects challenging since most are usually trained on large-size objects. These contain abundant information as they occupy a large number of pixels relative to the total image size. This fact is normally exploited by the model during training and inference processes. To dissect and understand this process, our approach systematically examines detectors' performances using two very distinct deep convolutional networks. The first is the single-stage YOLO V3 and the second is the double-stage Faster R-CNN. Specifically, our proposed method explores and visually illustrates the impact of feature extraction layers, number of anchor boxes, data augmentation, etc., utilising ideas from the field of explainable Artificial Intelligence (XAI). Our results, for example, show that multi-head YOLO V3 detectors trained using augmented data produce better performance even with a fewer number of anchor boxes. Moreover, robustness regarding the detector's ability to explain how a specific decision was reached is investigated using different explanation techniques. Finally, two new visualisation techniques are proposed, WS-Grad and Concat-Grad, for identifying explanation cues of different detectors. These are applied to specific object detection tasks to illustrate their reliability and transparency with respect to the decision process. It is shown that the proposed techniques can result in high resolution and comprehensive heatmaps of the image areas, significantly affecting detector decisions as compared to the state-of-the-art techniques tested.

A pixel-wise annotated dataset of small overlooked indoor objects for semantic segmentation applications.

The purpose of the dataset is to provide annotated images for pixel classification tasks with application to powered wheelchair users. As some of the widely available datasets contain only general objects, we introduced this dataset to cover the missing pieces, which can be considered as application-specific objects. However, these objects of interest are not only important for powered wheelchair users but also for indoor navigation and environmental understanding in general. For example, indoor assistive and service robots need to comprehend their surroundings to ease navigation and interaction with different size objects. The proposed dataset is recorded using a camera installed on a powered wheelchair. The camera is installed beneath the joystick so that it can have a clear vision with no obstructions from the user's body or legs. The powered wheelchair is then driven through the corridors of the indoor environment, and a one-minute video is recorded. The collected video is annotated on the pixel level for semantic segmentation (pixel classification) tasks. Pixels of different objects are annotated using MATLAB software. The dataset has various object sizes (small, medium, and large), which can explain the variation of the pixel's distribution in the dataset. Usually, Deep Convolutional Neural Networks (DCNNs) that perform well on large-size objects fail to produce accurate results on small-size objects. Whereas training a DCNN on a multi-size objects dataset can build more robust systems. Although the recorded objects are vital for many applications, we have included more images of different kinds of door handles with different angles, orientations, and illuminations as they are rare in the publicly available datasets. The proposed dataset has 1549 images and covers nine different classes. We used the dataset to train and test a semantic segmentation system that can aid and guide visually impaired users by providing visual cues. The dataset is made publicly available at this link.

Stakeholder views addressing the development and uptake of powered wheelchair assistive technology.

PURPOSE: The objective of this research is to identify stakeholder views with regard to the development of effective powered wheelchair assistive technologies more suited to the user and carer needs, whilst also meeting the requirements for other stakeholders, such that developers can be better guided towards producing solutions which have a better chance of getting to the market place and hence to the end user. METHOD: A questionnaire was designed to collect the views of all stakeholders and circulated to a statistically representative number of them. The question rating data were then checked for correlation between groups, and within groups, to establish validity. RESULTS: The 74 stakeholders across the eight classes who responded had a good correlation between each other, with a cross class "Pearson's correlation" ranging between 0.7 and 0.95, and the "Fleiss's Kappa reliability of agreement" within each class ranging between 0.07 and 0.36. CONCLUSIONS: This research has identified that all stakeholders should be involved in the development of the technology and that some may benefit in 'role-reversal' to help understand user problems and stakeholder concerns more clearly. Cost was a significant barrier to the uptake of appropriate technology, and training of users and carers was a major issue. Furthermore, development should not increase user isolation and the impact on the user must be monitored for 'quality of life'. Technical support and training should be given to the user and their carers, and equipment must be adaptive to meet the changing needs of the user. Implications for Rehabilitation Improved acceptance and use of technology by the user and their carers. Reduced rejection of appropriate provision. Improved mobility and interaction with others. Improved quality of life for users and carers.

Feature determination from powered wheelchair user joystick input characteristics for adapting driving assistance.

Background: Many powered wheelchair users find their medical condition and their ability to drive the wheelchair will change over time. In order to maintain their independent mobility, the powered chair will require adjustment over time to suit the user's needs, thus regular input from healthcare professionals is required. These limited resources can result in the user having to wait weeks for appointments, resulting in the user losing independent mobility, consequently affecting their quality of life and that of their family and carers. In order to provide an adaptive assistive driving system, a range of features need to be identified which are suitable for initial system setup and can automatically provide data for re-calibration over the long term. Methods: A questionnaire was designed to collect information from powered wheelchair users with regard to their symptoms and how they changed over time. Another group of volunteer participants were asked to drive a test platform and complete a course which represented manoeuvring in a very confined space as quickly as possible. Two of those participants were also monitored over a longer period in their normal home daily environment. Features, thought to be suitable, were examined using pattern recognition classifiers to determine their suitability for identifying the changing user input over time. Results: The results are not designed to provide absolute insight into the individual user behaviour, as no ground truth of their ability has been determined, they do nevertheless demonstrate the utility of the measured features to provide evidence of the users' changing ability over time whilst driving a powered wheelchair. Conclusions: Determining the driving features and adjustable elements provides the initial step towards developing an adaptable assistive technology for the user when the ground truths of the individual and their machine have been learned by a smart pattern recognition system.

Floor covering and surface identification for assistive mobile robotic real-time room localization application.

Assistive robotic applications require systems capable of interaction in the human world, a workspace which is highly dynamic and not always predictable. Mobile assistive devices face the additional and complex problem of when and if intervention should occur; therefore before any trajectory assistance is given, the robotic device must know where it is in real-time, without unnecessary disruption or delay to the user requirements. In this paper, we demonstrate a novel robust method for determining room identification from floor features in a real-time computational frame for autonomous and assistive robotics in the human environment. We utilize two inexpensive sensors: an optical mouse sensor for straightforward and rapid, texture or pattern sampling, and a four color photodiode light sensor for fast color determination. We show how data relating floor texture and color obtained from typical dynamic human environments, using these two sensors, compares favorably with data obtained from a standard webcam. We show that suitable data can be extracted from these two sensors at a rate 16 times faster than a standard webcam, and that these data are in a form which can be rapidly processed using readily available classification techniques, suitable for real-time system application. We achieved a 95% correct classification accuracy identifying 133 rooms' flooring from 35 classes, suitable for fast coarse global room localization application, boundary crossing detection, and additionally some degree of surface type identification.

Human peripheral blood monocytes express protease receptor-2 and respond to receptor activation by production of IL-6, IL-8, and IL-1{beta}.

Protease-activated receptor-2 (PAR-2) belongs to a family of G-coupled receptors activated by proteolytic cleavage to reveal a tethered ligand. PAR-2 is activated by trypsin and trypsin-like serine proteases and experimentally, by receptor-activating peptides (APs), which mimic the tethered ligand. PAR-2 has recently been implicated in proinflammatory immune responses. For example, PAR-2(-/-) mice exhibit markedly diminished contact hypersensitivity reactions and are completely resistant to adjuvant-induced arthritis. The present study shows that human blood monocytes express low-level cell-surface PAR-2 ex vivo, which is up-regulated upon cell purification by the mobilization of intracellular stores of PAR-2 protein. PAR-2 expression is also present on monocyte-derived macrophages, but only a small proportion of monocyte-derived dendritic cells (DC) is PAR-2(+), and blood DC are PAR(-). Freshly isolated monocytes responded to the PAR-2 AP ASKH 95 (2-furoyl-LIGKV-OH) with the generation of a calcium flux and production of interleukin (IL)-1beta, IL-6, and IL-8. The results presented thus suggest that PAR-2 contributes to inflammatory responses by inducing the production of proinflammatory cytokines in peripheral blood monocytes.

Fuzzy connectives as a combination tool in a hybrid multi-neural system.

The set of fuzzy connectives can be seen as an important combination tool, such as in combining the antecedent sets of the rules, in multi-criteria decision making and in combining the outputs of neural classifiers in a multi-neural system. This papers investigates the performance of some fuzzy combination schemes applied to a multi hybrid neural system which is composed of neural and fuzzy neural networks. An empirical evaluation in a handwritten numeral recognition task is used to investigate the performance of the presented fuzzy methods with some existing combination methods.

Tobacco dependence in a UK Bangladeshi female population: a cross-sectional study.

The objective of this study was to establish the prevalence of paan chewing with tobacco by UK-resident Bangladeshi women and the extent to which they manifest nicotine dependence. The cross-sectional study was conducted at two local authority housing estates in Tower Hamlets, London. Participants were 242 Bangladeshi women, selected at random from the current electoral register, who supplied a saliva sample for cotinine and an expired air sample for carbon monoxide analysis. They also participated in a structured interview assessing knowledge, attitudes and behavior with respect to tobacco use. Main outcome measures were data on tobacco use and nicotine dependence, assessed by questionnaire and intake measures. The population prevalence of chewing paan quid with tobacco was 48.5% (95% confidence interval, CI 42.01-54.98%), while 4% (95% CI 2.05-7.41%) smoked cigarettes. Higher mean salivary cotinine scores were associated with greater consumption frequency and use of leaf tobacco in the quid. Above-average nicotine dependence was associated with chewing paan quid with tobacco within 1 h of waking (OR = 4.02, p = 0.03, 95% CI 1.08-14.94) and the use of leaf rather than processed tobacco (OR = 3.91, p = 0.025, 95% CI 1.19-12.81). Smoking prevalence is low, but the prevalence of paan quid with tobacco chewing is high in this sample of Bangladeshi women. Cotinine concentration appears to be a reliable indicator of levels of nicotine dependence among paan quid with tobacco chewers. Questionnaire-derived items can be used to identify those with above-average levels of nicotine dependence.