Nowadays, the scale of Money Laundering is difficult to estimate in the UK and elsewhere. Proceeds of crimes might be transferred using the available business infrastructure offered by banks, and this is a considerable problem. This paper outlines a novel scheme that allows banks to share information leading to Money Laundering (ML) detection all the while preserving confidentiality and integrity. The main contribution is the overall architecture that aims to improve ML detection by getting other banks to collaborate. In order to get other banks to co-operate, a primary directive of preserving privacy is enforced throughout the framework. The proposed scheme has two particular aspects, one of which is the application of encrypted data used in machine learning for ML detection. Another feature is using secret sharing as a collaborative element in this context. These aspects are found in the three phases of the framework: Signalling to the Auditor, ML Detection and finally Suspicious Activity Report (SAR) Feedback.

Door phone systems, allowing occupants of a building to communicate with visitors at the door, have evolved over the years, with the current advancements being a fully internet protocol (IP) based solution. In order to adopt newer IP based solutions, current analogue systems can be replaced, yet this may be costly and cumbersome, especially in a conventional multi-occupant building. We therefore propose an architecture which supports current analogue door phone systems, and also provides IP based functionality. We have implemented the proposed architecture based on SIP, WebRTC and an IoT gateway system connected to the multi-occupant conventional video door phone system.

Free, confidential and uncensored communication between two individuals is an essential requirement for the modern, digital age. Social communication and interaction between users of Online Social Networks (OSNs) enjoy certain security properties, but true end-to-end security cannot be achieved. In this chapter, we devise a secure channel between two OSN Friends using social messaging. Confidentiality is achieved through undetectable communication based on distributed low-entropy steganography, thus implementing a security control for message content vis-a-vis an untrusted OSN provider. A prototype solution, in the form of an Android mobile app, is implemented. This uses two channels of social messaging (Twitter and Google+) to demonstrate the feasibility of the proposed approach. The source code is publicly available, thus making it suitable for use by individuals (and organisations) as a cost-free, secure communication tool.

This paper presents a method for pedestrian re-identification, with two novel contributions. Firstly, each element in the target population is classified into one of n categories, using the expected accuracy of the re-identification estimate for this element. A metric for each category is separately trained using a standard (Local Fisher) method. To process a test set, each element is classified into one of the categories, and the corresponding metric is selected and used. The second contribution is the proposal to use a symmetrised distance measure. A standard procedure is to learn a metric using one set as the probe and the other set as the gallery. This paper generalises that procedure by reversing the labels to learn a different metric, and uses a linear (symmetrised) combination of the two. This can be applied in cases for which there are two distinct sets of observations, i.e. from two cameras, e.g. VIPER. Using this publicly available dataset, it is demonstrated how these contributions result in improved re-identification performance.

This paper presents a novel approach for pedestrian detection using oriented line scans of gradients computed from a gray level image. Three feature types are proposed that can be generated easily from oriented gradients and an effective use of integral lines and integral images. A scalable cascaded classifier is built by combining oriented gradients with the oriented line scan features in a boosting framework. The detector's performance is comparable to the state of the art results and achieves about 3 to 5 fps on 320 x 240 resolution images making the proposed method suitable for real time applications. Detector performance is also represented as PUR, percentage of uncertainty removed.

This paper presents the application of fusion methods to a visual surveillance scenario. The range of relevant features for re-identifying vehicles is discussed, along with the methods for fusing probabilistic estimates derived from these estimates. In particular, two statistical parametric fusion methods are considered: Bayesian Networks and the Dempster Shafer approach. The main contribution of this paper is the development of a metric to allow direct comparison of the benefits of the two methods. This is achieved by generalising the Kelly betting strategy to accommodate a variable total stake for each sample, subject to a fixed expected (mean) stake. This metric provides a method to quantify the extra information provided by the Dempster-Shafer method, in comparison to a Bayesian Fusion approach.

A novel tracklet association framework is introduced to perform robust online re-identification of pedestrians in crowded scenes recorded by a single camera. Recent advances in multi-target tracking allow the generation of longer tracks, but problems of fragmentation and identity switching remain, due to occlusions and interactions between subjects. To address these issues, a discriminative and efficient descriptor is proposed to represent a tracklet as a bag of independent motion signatures using spatio-temporal histograms of oriented gradients. Due to the significant temporal variations of these features, they are generated only at automatically identified key poses that capture the essence of its appearance and motion. As a consequence, the re-identification involves only the most appropriate features in the bag at given time. The superiority of the methodology is demonstrated on two publicly available datasets achieving accuracy over 90

Plant species classification using leaf samples is a challenging and important problem to solve. This paper introduces a new data set of sixteen samples each of one-hundred plant species; and describes a method designed to work in conditions of small training set size and possibly incomplete extraction of features. This motivates a separate processing of three feature types: shape, texture, and margin; combined using a probabilistic framework. The texture and margin features use histogram accumulation, while a normalised description of contour is used for the shape. Two previously published methods are used to generate separate posterior probability vectors for each feature, using data associated with the k-Nearest Neighbour apparatus. The combined posterior estimates produce the final classification (where missing features could be omitted). We show that both density estimators achieved a 96% mean accuracy of classification when combining the three features in this way (training on 15 samples with unseen cross validation). In addition, the framework can provide an upper bound on the Bayes Risk of the classification problem, and thereby assess the accuracy of the density estimators. Lastly, the high performance of the method is demonstrated for small training set sizes: 91% accuracy is observed with only four training samples.

Metric learning methods, for person re-identification, estimate a scaling for distances in a vector space that is optimized for picking out observations of the same individual. This paper presents a novel approach to the pedestrian re-identification problem that uses metric learning to improve the state-of-the-art performance on standard public datasets. Very high dimensional features are extracted from the source color image. A first processing stage performs unsupervised PCA dimensionality reduction, constrained to maintain the redundancy in color-space representation. A second stage further reduces the dimensionality, using a Local Fisher Discriminant Analysis defined by a training set. A regularization step is introduced to avoid singular matrices during this stage. The experiments conducted on three publicly available datasets confirm that the proposed method outperforms the state-of-the-art performance, including all other known metric learning methods. Further-more, the method is an effective way to process observations comprising multiple shots, and is non-iterative: the computation times are relatively modest. Finally, a novel statistic is derived to characterize the Match Characteristic: the normalized entropy reduction can be used to define the 'Proportion of Uncertainty Removed' (PUR). This measure is invariant to test set size and provides an intuitive indication of performance.

This paper presents a method for localising and recognising vehicle manufacturer logos in both the front and rear views. The method assumes that the vehicle registration plate is visible and an estimate of its location is available. Features are constructed out of local histograms of gradients, in both conventional and hierarchical arrangements. The dimensionality of these vectors is then reduced using unsupervised PCA and, subsequently a supervised method based on Local Fisher Discriminant Analysis. This then provides a suitable metric for logo detection, localisation and multi-class classification. On a test set of data captured from a medium range CCTV camera, with five different manufacturers' logos, the proposed method provided a correct logo localisation rate of 97 % and a correct logo classification rate of 87.6%.

K-nearest-neighbours is a simple classifier, and with increasing size of training set, the accuracy of its class predictions can be made asymptotic to the upper bound.Probabilistic classifications can also be generated: the accuracy of a simple proportional scheme will also be asymptotic to the upper bound, in the case of large training sets.Outside this limit, this and other existing schemes make ineffective use of the available information: this paper proposes a more accurate method, that improves the state-of-the-art performance, evaluated on several public data sets.Criteria such as the degree of unanimity among the neighbours, the observed rank of the correct class, and the intra-class confusion matrix can be used to tabulate the observed classification accuracy within the (cross-validated) training set.These tables can then be used to make probabilistic class predictions for the previously unseen test set, used to evaluate the novel and previous methods in two ways: i) mean a posteriori probability and ii) accuracy of the discrete prediction obtained from integrating the probabilistic estimates from independent sources.The proposed method performs particularly well in the limit of small training set sizes.

A novel co-training framework is proposed for object orientation estimation in a multi-camera network environment. The model is initialised using a small labelled dataset and then iteratively boosted using large amount of unlabelled data which are generated automatically from videos. This optimisation process is guided by pairwise constraints of known orientation difference between two views of an object. The introduced methodology is combined with Support Vector Machine and Expectation-Maximization algorithm. The thorough experimental evaluation using 3 datasets of football players, pedestrians and cars confirms the superiority of the boosted models for a robust orientation estimation.

Protecting citizens' privacy is a critical issue in the nowadays society. As data collection capabilities increase with the introduction of new technologies for automatic data capturing, new threats related to privacy protection need to be faced. In this field, video surveillance represents a critical technology: the existence of large amounts of stored personal data increases the probability of personal data misuse. In this paper, a technology that tackles the complex challenge of balancing security and privacy in video-surveillance is presented. The technology has potential for a greater social acceptance of video surveillance technology, due to the inclusion of the value of privacy alongside the enhanced security.

The Image Library for Intelligent Detection Systems (i-LIDS) provides benchmark surveillance datasets for analytics systems. This paper proposes a methodology to investigate the effect of compression and frame-rate reduction, and to recommend an appropriate suite of degraded datasets for public release. The library consists of six scenarios, including Sterile Zone (SZ) and Parked Vehicle (PV), which are investigated using two different compression algorithms (H. 264 and JPEG) and a number of detection systems. PV has higher spatio-temporal complexity than the SZ. Compression performance is dependent on scene content hence PV will require larger bit-streams in comparison with SZ, for any given distortion rate. The study includes both industry standard algorithms (for transmission) and CCTV recorders (for storage). CCTV recorders generally use proprietary formats, which may significantly affect the visual information. Encoding standards such as H. 264 and JPEG use the Discrete Cosine Transform (DCT) technique, which introduces blocking artefacts. The H. 264 compression algorithm follows a hybrid predictive coding approach to achieve high compression gains, exploiting both spatial and temporal redundancy. The highly predictive approach of H. 264 may introduce more artefacts resulting in a greater effect on the performance of analytics systems than JPEG. The paper describes the two main components of the proposed methodology to measure the effect of degradation on analytics performance. Firstly, the standard tests, using the 'f-measure' to evaluate the performance on a range of degraded video sets. Secondly, the characterisation of the datasets, using quantification of scene features, defined using image processing techniques. This characterization permits an analysis of the points of failure introduced by the video degradation.

The significant progress in visual surveillance has been motivated by the need to emulate some of the human ability to monitor activity in human-made environments, particularly in the contexts of security and safety. The rapid rise in numbers of cameras installed in public and private places makes such automation desirable, at least to reduce CCTV workload. Real-world applications of visual surveillance impose the need of robust real-time solutions, able to deal with a wide range of circumstances and environmental conditions. Conventional approaches work based on what has become known as motion (or change) detection followed by tracking (in single or multiple camera systems). Objects of interest are represented by rectangular blobs and decisions on whether something might be interesting are made on rules or learned patterns of presence and trajectories of such blobs. There is growing interest in looking ‘inside the box’ for applications that are concerned with detailed human activity recognition and with robust detection of people even when image backgrounds change, as is the case of a moving camera. In this study, the authors consider the general problem of robust pedestrian detection irrespective of background, reviewing the state of the art, showing some representative results and suggesting ways forward.

Computer programming is a highly practical subject and it is essential that those new to the discipline engage in hands-on experimentation as part of the learning process. However, when faced with large cohorts and an increasing demand for distance and student flexible learning, incorporating this into a programming course can be difficult. There is a dynamic that exists between tutor and student in a real-world programming workshop session that is not easily replicated online. In this paper we describe an online learning environment that begins to create an analogue of this dynamic and its successful integration into an undergraduate programming module. Ultimately, the potential exists to not only improve the student learning experience but also investigate and inform programming pedagogy itself.

This paper presents a new tracking algorithm to solve on-line the 'Tag and Track' problem in a crowded scene with a network of CCTV Pan, Tilt and Zoom (PTZ) cameras. The dataset is very challenging as the non-overlapping cameras exhibit pan tilt and zoom motions, both smoothly and abruptly. Therefore a tracking-by-detection approach is combined with a re-identification method based on appearance features to solve the re-acquisition problem between non overlapping camera views and crowds occlusions. However, conventional re-identification techniques of multi target trackers, which consist of learning an online appearance model to differentiate the target of interest from other people in the scene, are not suitable for this scenario because the tagged pedestrian moves in an environment where pedestrians walking with them are constantly changing. Therefore, a novel multiple shots re-identification technique is proposed which combines a standard single shot re-identification, based on offline training to recognize humans from different views, with a Dynamic Time Warping (DTW) distance.