In this paper we advance the state-of-the-art for crowd counting in high density scenes by further exploring idea of a fully convolutional model introduced (Zhang et al., 2016).Producing an accurate and robust count estimator using computer vision techniques has attracted significant research interest recent years.Applications systems exist many diverse areas including city planning, retail, course general public safety.Developing highly generalised that can be deployed any surveillance...

In this paper we propose ResnetCrowd, a deep residual architecture for simultaneous crowd counting, violent behaviour detection and density level classification. To train evaluate the proposed multi-objective technique, new 100 image dataset referred to as Multi Task Crowd is constructed. This first computer vision fully annotated Our experiments show that multi-task approach boosts individual task performance all tasks most notably which receives 9% boost in ROC curve AUC (Area under...

In this paper we propose a technique to adapt convolutional neural network (CNN) based object counter additional visual domains and types while still preserving the original counting function. Domain-specific normalisation scaling operators are trained allow model adjust statistical distributions of various domains. The developed adaptation is used produce singular patch-based regressor capable including people, vehicles, cell nuclei wildlife. As part study challenging new dataset in context...

This paper presents a new approach to crowd behaviour anomaly detection that uses set of efficiently computed, easily interpretable, scene-level holistic features. low-dimensional descriptor combines two features from the literature: collectiveness and conflict, with newly developed features: mean motion speed formulation density. Two different approaches are investigated using these When only normal training data is available we use Gaussian Mixture Model (GMM) for outlier detection. both...

Hand-shape recognition is an important problem in computer vision with significant societal impact. In this work, we introduce a new image dataset for Irish Sign Language (ISL) and compare between two approaches. The was collected by filming human subjects performing ISL hand-shapes movements. Then, extracted frames from the videos. This produced total of 52,688 images 23 common hand- shapes ISL. Afterwards, filter redundant iterative selection process that selects which keep diverse. For...

Unlike land, the oceans, although covering more than 70% of planet, are largely unexplored. Global fisheries resources central to sustainability and quality life on earth but under threat from climate change, ocean acidification over consumption. One way analyze these marine resource is through remote underwater surveying. However, sheer volume recorded data often make classification analyses difficult, time consuming intensive. Recent developments in machine learning (ML) have shown...

Semantic annotation of digital objects within large multimedia collections is a difficult and challenging task. We describe method for semi-automatic images apply it to evaluate on pancreatic cells. By comparing the performance this approach in cell domain with previous results fuel domain, we aim determine characteristics which indicate that will or not work domain. conclude by describing types domains can expect satisfactory approach.

Video analytics has a key role to play in smart cities and connected community applications such as crowd counting, activity detection, event classification, traffic counting etc. Using cloud-centric approach where data is funnelled central processor presents number of problems available bandwidth, real-time responsiveness personal privacy issues. With the development edge computing, new paradigm for management emerging. Raw video feeds can be pre-processed at point capture while integration...

The popularity of deep learning has increased tremendously in recent years due to its ability efficiently solve complex tasks challenging areas such as computer vision and language processing. Despite this success, low-level neural activity reproduced by Deep Neural Networks (DNNs) generates extremely rich representations the data. These are difficult characterise cannot be directly used understand decision process. In paper we build upon our exploratory work where introduced concept a...

Diminishing reserves of cheap, recoverable crude oil, together with increasing tensions between the Middle East and West, will likely threaten our access to affordable oil in future. Alternative fossil fuels such as coal, tar sand, heavy only worsen global warming. Hydrogen, however, is plentiful clean, stores energy more effectively than batteries, burns twice efficiently a fuel cell gasoline does an internal-combustion engine, leaves water behind, can power cars. But many challenges remain...

This paper presents work on integrating multiple computer vision-based approaches to surveillance video analysis support user retrieval of segments showing human activities. Applied vision using real-world data is an extremely challenging research problem, independently any information (IR) issues. Here we describe the issues faced in developing both generic and specific tools how they were integrated for use new TRECVid interactive event detection task. We present interaction paradigm...

Over the years, paradigm of medical image analysis has shifted from manual expertise to automated systems, often using deep learning (DL) systems. The performance algorithms is highly dependent on data quality. Particularly for domain, it an important aspect as very sensitive quality and poor can lead misdiagnosis. To improve diagnostic performance, research been done both in complex DL architectures improving dataset static hyperparameters. However, still constrained due overfitting...

Detailed, consistent semantic annotation of large collections multimedia data is difficult and time- consuming. In domains such as eScience, digital curation industrial monitoring, fine-grained high- quality labeling regions enables advanced querying, analysis aggregation supports collaborative research. Manual inefficient too subjective to be a viable solution. Automatic solutions are often highly domain or application specific, require volumes annotated training corpi and, if using `black...

Understanding hydrological processes in large, open areas, such as catchments, and further modelling these are still research questions. The system proposed this work provides an automatic end-to-end pipeline from data collection to information extraction that can potentially assist hydrologists better understand the using a data-driven approach. In work, performance of low-cost off-the-shelf self contained sensor unit, which was originally designed used monitor liquid levels, AdBlue, fuel,...

Rapid urbanization has brought about an influx of people to cities, tipping the scale between urban and rural living. Population predictions estimate that 64% global population will reside in cities by 2050. To meet growing resource needs, improve management, reduce complexities, eliminate unnecessary costs while enhancing quality life citizens, are increasingly exploring open innovation frameworks smart city initiatives target priority areas including transportation, sustainability,...