Recognising human attributes from surveillance footage is widely studied for attribute-based re-identification. However, most works assume coarse, expertly-defined categories, ineffective in describing challenging images. Such brittle representations are limited in descriminitive power and hamper the efficacy of learnt estimators. We aim to discover more relevant and precise subject descriptions, improving image retrieval and closing the semantic gap. Inspired by fine-grained and relative attributes, we introduce super-fine attributes, which now describe multiple, integral concepts of a single trait as multi-dimensional perceptual coordinates. Crowd prototyping facilitates efficient crowdsourcing of super-fine labels by pre-discovering salient perceptual concepts for prototype matching. We re-annotate gender, age and ethnicity traits from PETA, a highly diverse (19K instances, 8.7K identities) amalgamation of 10 re-id datasets including VIPER, CUHK and TownCentre. Employing joint attribute regression with the ResNet-152 CNN, we demonstrate substantially improved ranked retrieval performance with super-fine attributes in comparison to conventional binary labels, reporting up to a 11.2 and 14.8 percent mAP improvement for gender and age, further surpassed by ethnicity. We also find our 3 super-fine traits to outperform 35 binary attributes by 6.5 percent mAP for subject retrieval in a challenging zero-shot identification scenario.

In order to improve recognition performance, fusion has become a key technique in the recent years. Compared with single-mode biometrics, the recognition rate of multi-modal biometric systems is improved and the final decision is more confident. This paper introduces a novel joint density distribution based rank-score fusion strategy that combines rank and score information. Recognition at a distance has only recently been of interest in soft biometrics. We create a new soft biometric database containing the human face, body and clothing attributes at three different distances to investigate the influence by distance on soft biometric fusion. A comparative study about our method and other state of the art rank level and score level fusion methods are also conducted in this paper. The experiments are performed using a soft biometric database we created. The results demonstrate the recognition performance is significantly improved by our proposed method.

In this paper we introduce an algorithm for 3D motion estimation in point clouds that is based on Chasles' kinematic theorem. The proposed algorithm estimates 3D motion parameters directly from the data by exploiting the geometry of rigid transformation using an evidence gathering technique in a Hough-voting-like approach. The algorithm provides an alternative to the feature description and matching pipelines commonly used by numerous 3D object recognition and registration algorithms, as it does not involve keypoint detection and feature descriptor computation and matching. To the best of our knowledge, this is the first research to use kinematics theorems in an evidence gathering framework for motion estimation and surface matching without the use of any given correspondences. Moreover, we propose a method for voting for 3D motion parameters using a one-dimensional accumulator space, which enables voting for motion parameters more efficiently than other methods that use up to 7-dimensional accumulator spaces.

Automatically describing pedestrians in surveillance footage is crucial to facilitate human accessible solutions for suspect identification. We aim to identify pedestrians based solely on human description, by automatically retrieving semantic attributes from surveillance images, alleviating exhaustive label annotation. This work unites a deep learning solution with relative soft biometric labels, to accurately retrieve more discriminative image attributes. We propose a Semantic Retrieval Convolutional Neural Network to investigate automatic retrieval of three soft biometric modalities, across a number of `closed-world' and `open-world' re-identification scenarios. Findings suggest that relative-continuous labels are more accurately predicted than absolute-binary and relative-binary labels, improving semantic identification in every scenario. Furthermore, we demonstrate a top rank-1 improvement of 23.2% and 26.3% over a traditional, baseline retrieval approach, in one-shot and multi-shot re-identification scenarios respectively.

Much progress has been made recently in the development of 3D acquisition technologies, which increased the availability of low-cost 3D sensors, such as the Microsoft Kinect. This promotes a wide variety of computer vision applications needing object recognition and 3D reconstruction. We present a novel algorithm for full 3D reconstruction of unknown rotating objects in 2.5D point cloud sequences, such as those generated by 3D sensors. Our algorithm incorporates structural and temporal motion information to build 3D models of moving objects and is based on motion compensated temporal accumulation. The proposed algorithm requires only the fixed centre or axis of rotation, unlike other 3D reconstruction methods, it does not require key point detection, feature description, correspondence matching, provided object models or any geometric information about the object. Moreover, our algorithm integrally estimates the best rigid transformation parameters for registration, applies surface resembling, reduces noise and estimates the optimum angular velocity of the rotating object.

Many studies have shown that it is possible to recognize people by the way they walk. However, there are a number of covariate factors that affect recognition performance. The time between capturing the gallery and the probe has been reported to affect recognition the most. To date, no study has isolated the effect of time, irrespective of other covariates. Here, we present the first principled study that examines the effect of elapsed time on gait recognition. Using empirical evidence we show for the first time that elapsed time does not affect recognition significantly in the short-medium term. This finding challenges the existing view in the literature that time significantly affects gait recognition. We employ existing gait representations on a novel dataset captured specifically for this study. By controlling the clothing worn by the subjects and the environment, a Correct Classification Rate (CCR) of 95% has been achieved over the longest time period yet considered for gait on the largest ever temporal dataset. Our results show that gait can be used as a reliable biometric over time and at a distance if we were able to control all other factors such as clothing, footwear etc. We have also investigated the effect of different type of clothes, variations in speed and footwear on the recognition performance. The purpose of these experiments is to provide an indication of why previous studies (employing the same techniques as this study) have achieved significantly lower recognition performance over time. Our experimental results show that clothing and other covariates have been confused with elapsed time previously in the literature. We have demonstrated that clothing drastically affects the recognition performance regardless of elapsed time and significantly more than any of the other covariates that we have considered here.

In this paper, we explore a new approach for enriching the HoG method for pedestrian detection in an unconstrained outdoor environment. The proposed algorithm is based on using gait motion since the rhythmic footprint pattern for walking people is considered the stable and characteristic feature for the detection of walking people. The novelty of our approach is motivated by the latest research for people identification using gait. The experimental results confirmed the robustness of our method to enhance HoG to detect walking people as well as to discriminate between single walking subject, groups of people and vehicles with a detection rate of 100%. Furthermore, the results revealed the potential of our method to be used in visual surveillance systems for identity tracking over different camera views.

We present a new method for viewpoint independent gait biometrics. The system relies on a single camera, does not require camera calibration, and works with a wide range of camera views. This is achieved by a formulation where the gait is self-calibrating. These properties make the proposed method particularly suitable for identification by gait, where the advantages of completely unobtrusiveness, remoteness, and covertness of the biometric system preclude the availability of camera information and specific walking directions. The approach has been assessed for feature extraction and recognition capabilities on the SOTON gait database and then evaluated on a multiview database to establish recognition capability with respect to view invariance. Moreover, tests on the multiview CASIA-B database, composed of more than 2270 video sequences with 65 different subjects walking freely along different walking directions, have been performed. The obtained results show that human identification by gait can be achieved without any knowledge of internal or external camera parameters with a mean correct classification rate of 73.6% across all views using purely dynamic gait features. The performance of the proposed method is particularly encouraging for application in surveillance scenarios.

We present a new method for viewpoint independent markerless gait analysis. The system uses a single camera, does not require camera calibration and works with a wide range of directions of walking. These properties make the proposed method particularly suitable for identification by gait, where the advantages of completely unobtrusiveness, remoteness and covertness of the biometric system preclude the availability of camera information and use of marker based technology. Tests on more than 200 video sequences with subjects walking freely along different walking directions have been performed. The obtained results show that markerless gait analysis can be achieved without any knowledge of internal or external camera parameters and that the obtained data that can be used for gait biometrics purposes. The performance of the proposed method is particularly encouraging for its appliance in surveillance scenarios.

The development of large scale biometric systems requires experiments to be performed on large amounts of data. Existing capture systems are designed for fixed experiments and are not easily scalable. In this scenario even the addition of extra data is difficult. We developed a prototype biometric tunnel for the capture of non-contact biometrics. It is self contained and autonomous. Such a configuration is ideal for building access or deployment in secure environments. The tunnel captures cropped images of the subject's face and performs a 3D reconstruction of the person's motion which is used to extract gait information. Interaction between the various parts of the system is performed via the use of an agent framework. The design of this system is a trade-off between parallel and serial processing due to various hardware bottlenecks. When tested on a small population the extracted features have been shown to be potent for recognition. We currently achieve a moderate throughput of approximate 15 subjects an hour and hope to improve this in the future as the prototype becomes more complete

Recognizing people by gait has a unique advantage over other biometrics: it has potential for use at a distance when other biometrics might be at too low a resolution, or might be obscured. The current state of the art can achieve over 90% identification rate under situations where the training and test data are captured under similar conditions, while recognition rates with change of clothing, shoe, surface, illumination, and pose usually decrease performance and are the subject of much of the current study. Recognition can be achieved on outdoor data with uncontrolled illumination and at a distance when other biometrics could not be used. We shall show how this position has been achieved, covering most approaches to recognition by gait and the databases on which performance has been evaluated. We shall describe the context of these approaches, show how recognition by gait can be achieved and how current limits on performance are understood. We shall describe results on the most popular database, showing how recognition can handle some of the covariates that can affect recognition. We shall also investigate the supporting literature for this research, since the notion that people can be recognized by gait has support not only in medicine and biomedicine, and also in literature and psychology and other areas. In this way, we shall show that this new biometric has capability and research and application potential in other domains

Recently, a lot of research has been conducted into the usefulness of gait for identification at a distance. Since the gait of a person is readily identified when extracted from a canonical side view, most algorithms work with the premise that the motion is frontoparallel in nature, or require some knowledge of the camera calibration. Realistically people always walk along different trajectories to the camera. In this paper we show that gait has sufficient properties that allows us to exploit the structure of articulated motion within single view sequences, in order to remove the unknown subject pose and reconstruct the underlying gait signature, with no prior knowledge of the camera calibration.

Large arrays of cameras are increasingly being employed for producing high quality image sequences needed for motion analysis research. This leads to the logistical problem with coordination and control of a large number of cameras. In this paper, we used a lightweight multi-agent system for coordinating such camera arrays. The agent framework provides more than a remote sensor access API. It allows reconfigurable and transparent access to cameras, as well as software agents capable of intelligent processing. Furthermore, it eases maintenance by encouraging code reuse. Additionally, our agent system includes an automatic discovery mechanism at startup, and multiple language bindings. Performance tests showed the lightweight nature of the framework while validating its correctness and scalability. Two different camera agents were implemented to provide access to a large array of distributed cameras. Correct operation of these camera agents was confirmed via several image processing agents.

This paper describes a robust algorithm for arbitrary object tracking in long image sequences. This technique extends the dynamic Hough transform proposed in our earlier work to detect arbitrary shapes undergoing affine motion. The proposed tracking algorithm processes the whole image sequence globally. First, the object boundary is represented in lookup-table form, and we then perform an operation that estimates the energy of the motion trajectory in the parameter space. We assign an extra term in our cost function to incorporate smoothness of deformation. The object is actually rigid, so by 'deformation' we mean changes due to rotation or scaling of the object. There is no need for training or initialization, and an efficient implementation can be achieved with coarse-to-fine dynamic programming and pruning. The method, because of its evidence-based nature, is robust under noise and occlusion.

This paper described a robust algorithm for arbitrary object tracking in long image sequences. This technique extends the dynamic Hough transform proposed in our earlier work to detect arbitrary shapes undergoing affine motion. The proposed tracking algorithm processes the whole image sequence globally. First, the object boundary is represented in lookup-table form, and we then perform an operation that estimates the energy of the motion trajectory in the parameter space. We assign an extra term in our cost function to incorporate smoothness of deformation. The object is actually rigid, so by 'deformation' we mean changes due to rotation or scaling of the object. There is no need for training or initialization, and an efficient implementation can be achieved with coarse-to-fine dynamic programming and pruning. The method, because of its evidence-based nature, is robust under noise and occlusion.

We describe a new method for automatic gait recognition based on analysing the symmetry of human motion using the Generalised Symmetry Operator. This approach is reinforced by the psychologists’ view that gait is a symmetrical pattern of motion and results show that gait can indeed be recognised by symmetry analysis.

Recent research using statistical moments to describe moving shapes through an image sequence has led to an interest in reconstructing moving shapes from their moment description. This paper discusses how the moment description through a series of frames might be used to predict missing or intermediate frames within a sequence. Additionally, this highlights generic aspects of moment reconstruction which rarely receive more than scant attention. The ideas presented use Zernike moments, although the general framework is applicable to all types of moments. We show how a moving human silhouette can be reconstructed with accuracy by interpolation from a moment history. 1

We show how the symmetry of motion can be extracted by using the generalised symmetry operator for analysing motion and for gait recognition. This operator, rather than relying on the borders of a shape or on general appearance, locates features by their symmetrical properties. This approach is reinforced by the view from psychology that human gait is a symmetrical pattern of motion, and by other works. We applied our new method to compare animal gait, and for recognition by gait. Results show that the symmetry properties of gait appear to be unique and can indeed be used for analysis and for recognition. We have so far achieved promising recognition rates of over 95%. Performance analysis also suggests that symmetry enjoys practical advantages such as relative immunity to noise with capability to handle occlusion and as such might prove suitable for applications like clip-database browsing.

Motion tracking is an important task in computer vision. A new technique, the dynamic velocity Hough transform (DVHT), for tracking of parametric objects is described that extends the velocity Hough transform (VHT) to cater for arbitrary motion. Like the VHT, the new technique processes the whole image sequence, gathering global evidence of motion and structure. However, we do not assume constant linear velocity but rather allow arbitrary velocity. The method tries to find an optimal, smooth trajectory in the parameter space with maximum energy, where the latter incorporates both the structure of the moving object and the smoothness of motion. The constrained optimisation problem is solved using a temporal (time-delay) dynamic programming algorithm. Tracking in noise is much superior to the standard Hough transform.

The objective in defining feature space is to reduce the dimension of the original pattern space yet maintaining discriminatory power for classification. To meet this objective in the context of ear and face biometrics a novel force field transformation has been developed in which the image is treated as an array of Gaussian attractors that act as the source of a force field. The directional properties of the force field are exploited to automatically locate a small number of potential energy wells and channels that form the basis of a characteristic feature vector. Here, we generalise the analysis, and the stock of applications.

The overall objective in de®ning feature space is to reduce the dimensionality of pattern space yet maintaining discriminatory power for classi®cation and invariant description.To meet this objective, in the context of ear biometrics, a novel force ®eld transformation has been developed in which the image is treated as an array of Gaussian attractors that act as the source of a force ®eld.The directional properties of the force ®eld are exploited to automatically locate the extrema of a small number of potential energy wells and associated potential channels.These form the basis of the ear description.This has been applied to a small database of ears and initial results show that the new approach has suitable performance attributes and shows promising results in automatic ear recognition.

We present a new evidence gathering based approach, aimed to extract moving articulated objects from a temporal sequence of images. The new technique is designed to enable the automated determination of parameters pertaining to human gait, with a view to possible use as a biometric for recognition purposes. The articulated line feature extraction technique, uses a genetic algorithm (GA) based implementation of the Velocity Hough Transform (VHT). Using a GA to perform a heuristic search of the parameter space, rather than an exhaustive one, overcomes the problems of computation time and memory requirements associated with the original approach. The new technique employs a parametric gait model consisting of a pair of articulated lines, jointed at the hip. Trials on real image sequences of pedestrians demonstrate that the approach is capable of locating and tracking a walking subject. Moreover, the technique is able to provide reasonable estimates of an individual’s gait cycle period and hip rotation patterns, which are pertinent to recognition. However, current levels of accuracy are insufficient for these purposes. Nevertheless, the results demonstrate that the articulated line feature extraction technique has potential for use as an automated gaitdata retrieval system. 1

We propose a new dynamic feature extraction technique for the analysis of temporal image sequences. Based on the Hough transform, the new technique includes motion in the evidence gathering process to detect moving parametric shapes. Experimental results on real and synthetic image sequences show that this new technique offers increased noise immunity and improves performance, especially in cases of partial occlusion. The new technique is extended to enable the identification of features exhibiting a time variant structure and applied to the task of extracting arterial wall data from a sequence of cross-sectional ultrasound images. Experimental results demonstrate that the approach can provide good results and highlights the performance of the new technique in noisy environments.

Magnetic Resonance Imaging(MRI) has been used to measure the shape of the vocal tract during speech in several recent studies. Its safety to the subject, high quality imaging of soft tissue, and the ability to select relatively thin imaging planes at any angle are significant advantages over other imaging methods used for speech research. The most significant disadvantage is the long exposure time. As a result most studies have focused on obtaining high-resolution images of the vocal tract volume for static sounds, such as vowels[1], fricatives[5,6], nasals, the closed phase of plosives[7] and liquids[3,7]. In this paper we will describe our method of obtaining MR images of a moving vocal tract in which we post-synchronize the MR data using a recorded speech signal and thus reconstruct the images without using the MR machine's built-in processing.

In order to model speech production for purposes such as articulatory synthesis, articulatory data must be acquired, preferably in a way that does not impede the speaker's ability to speak. A variety of techniques has been used, many of them, such as ultrasound, X-rays, or MRI, adapted from medical imaging. For externally accessible articulators such techniques are not appropriate, but the task of measuring the shape of a complex three-dimensional object is still difficult. For a static object such as a dental impression manual methods such as slicing a mold of the impression and measuring the slices with calipers can be quite accurate, though time-consuming. For a dynamic object such as the face shape during speech such manual methods cannot be used. A workable alternative has been developed at Grenoble that instead uses simultaneous video pictures of front and profile face. Blue lipstick is used on the subject's lips to provide a definite outline and provide maximum contrast with the teeth and tongue. The images must then be post-processed before parameters such as mouth area are extracted. In this paper we present a structured light system which uses a slide projector and a single video camera to acquire depth coordinates of a static or moving object. We discuss pilot experiments aimed at optimising its output and establishing its accuracy, and make some preliminary comparisons with the Grenoble double-video system. In the next section the structured light technique is summarised, focusing on the experimental constraints it imposes. The technique is described in more detail elsewhere. In Static measurements' its use on the static object of an EPG palate is described. In 'Dynamic measurements' its use in acquiring dynamic face shape data is described, focusing on methodology issues that arose in an extended recording session with a human speaker. In the discussion we describe the calibration procedure and the related tradeoffs, and finally we make some initial comparisons to the double-video system.