Supervised latent dirichlet allocation models for efficient activity representation

Local spatio-temporal features with a Bag-of-visual words model is a popular approach used in human action recognition. Bag-of-features methods suffer from several challenges such as extracting appropriate appearance and motion features from videos, converting extracted features appropriate for classification and designing a suitable classification framework. In this paper we address the problem of efficiently representing the extracted features for classification to improve the overall performance. We introduce two generative supervised topic models, maximum entropy discrimination LDA (MedLDA) and class- specific simplex LDA (css-LDA), to encode the raw features suitable for discriminative SVM based classification. Unsupervised LDA models disconnect topic discovery from the classification task, hence yield poor results compared to the baseline Bag-of-words framework. On the other hand supervised LDA techniques learn the topic structure by considering the class labels and improve the recognition accuracy significantly. MedLDA maximizes likelihood and within class margins using max-margin techniques and yields a sparse highly discriminative topic structure; while in css-LDA separate class specific topics are learned instead of common set of topics across the entire dataset. In our representation first topics are learned and then each video is represented as a topic proportion vector, i.e. it can be comparable to a histogram of topics. Finally SVM classification is done on the learned topic proportion vector. We demonstrate the efficiency of the above two representation techniques through the experiments carried out in two popular datasets. Experimental results demonstrate significantly improved performance compared to the baseline Bag-of-features framework which uses kmeans to construct histogram of words from the feature vectors.

Dynamics of contour, object and face processing in the human visual cortex

The neural basis of visual perception can be understood only when the sequence of cortical activity underlying successful recognition is known. The early steps in this processing chain, from retina to the primary visual cortex, are highly local, and the perception of more complex shapes requires integration of the local information. In Study I of this thesis, the progression from local to global visual analysis was assessed by recording cortical magnetoencephalographic (MEG) responses to arrays of elements that either did or did not form global contours. The results demonstrated two spatially and temporally distinct stages of processing: The first, emerging 70 ms after stimulus onset around the calcarine sulcus, was sensitive to local features only, whereas the second, starting at 130 ms across the occipital and posterior parietal cortices, reflected the global configuration. To explore the links between cortical activity and visual recognition, Studies II III presented subjects with recognition tasks of varying levels of difficulty. The occipito-temporal responses from 150 ms onwards were closely linked to recognition performance, in contrast to the 100-ms mid-occipital responses. The averaged responses increased gradually as a function of recognition performance, and further analysis (Study III) showed the single response strengths to be graded as well. Study IV addressed the attention dependence of the different processing stages: Occipito-temporal responses peaking around 150 ms depended on the content of the visual field (faces vs. houses), whereas the later and more sustained activity was strongly modulated by the observers attention. Hemodynamic responses paralleled the pattern of the more sustained electrophysiological responses. Study V assessed the temporal processing capacity of the human object recognition system. Above sufficient luminance, contrast and size of the object, the processing speed was not limited by such low-level factors. Taken together, these studies demonstrate several distinct stages in the cortical activation sequence underlying the object recognition chain, reflecting the level of feature integration, difficulty of recognition, and direction of attention.

Deep learning architecture for gesture recognition

Pós-graduação em Engenharia Mecânica - FEG

A Novel Prediction Method for Early Recognition of Global Human Behaviour in Image Sequences

Human behaviour recognition has been, and still remains, a challenging problem that involves different areas of computational intelligence. The automated understanding of people activities from video sequences is an open research topic in which the computer vision and pattern recognition areas have made big efforts. In this paper, the problem is studied from a prediction point of view. We propose a novel method able to early detect behaviour using a small portion of the input, in addition to the capabilities of it to predict behaviour from new inputs. Specifically, we propose a predictive method based on a simple representation of trajectories of a person in the scene which allows a high level understanding of the global human behaviour. The representation of the trajectory is used as a descriptor of the activity of the individual. The descriptors are used as a cue of a classification stage for pattern recognition purposes. Classifiers are trained using the trajectory representation of the complete sequence. However, partial sequences are processed to evaluate the early prediction capabilities having a specific observation time of the scene. The experiments have been carried out using the three different dataset of the CAVIAR database taken into account the behaviour of an individual. Additionally, different classic classifiers have been used for experimentation in order to evaluate the robustness of the proposal. Results confirm the high accuracy of the proposal on the early recognition of people behaviours.

Attentional shifts towards an expected visual target alter the level of alpha-band oscillatory activity in the human calcarine cortex

Neuronal operations associated with the top-down control process of shifting attention from one locus to another involve a network of cortical regions, and their influence is deemed fundamental to visual perception. However, the extent and nature of these operations within primary visual areas are unknown. In this paper, we used magnetoencephalography (MEG) in combination with magnetic resonance imaging (MRI) to determine whether, prior to the onset of a visual stimulus, neuronal activity within early visual cortex is affected by covert attentional shifts. Time/frequency analyses were used to identify the nature of this activity. Our results show that shifting attention towards an expected visual target results in a late-onset (600 ms postcue onset) depression of alpha activity which persists until the appearance of the target. Independent component analysis (ICA) and dipolar source modeling confirmed that the neuronal changes we observed originated from within the calcarine cortex. Our results further show that the amplitude changes in alpha activity were induced not evoked (i.e., not phase-locked to the cued attentional task). We argue that the decrease in alpha prior to the onset of the target may serve to prime the early visual cortex for incoming sensory information. We conclude that attentional shifts affect activity within the human calcarine cortex by altering the amplitude of spontaneous alpha rhythms and that subsequent modulation of visual input with attentional engagement follows as a consequence of these localized changes in oscillatory activity. © 2005 Elsevier B.V. All rights reserved.

Adaptive unsupervised learning of human actions

Automatic detection of suspicious activities in CCTV camera feeds is crucial to the success of video surveillance systems. Such a capability can help transform the dumb CCTV cameras into smart surveillance tools for fighting crime and terror. Learning and classification of basic human actions is a precursor to detecting suspicious activities. Most of the current approaches rely on a non-realistic assumption that a complete dataset of normal human actions is available. This paper presents a different approach to deal with the problem of understanding human actions in video when no prior information is available. This is achieved by working with an incomplete dataset of basic actions which are continuously updated. Initially, all video segments are represented by Bags-Of-Words (BOW) method using only Term Frequency-Inverse Document Frequency (TF-IDF) features. Then, a data-stream clustering algorithm is applied for updating the system's knowledge from the incoming video feeds. Finally, all the actions are classified into different sets. Experiments and comparisons are conducted on the well known Weizmann and KTH datasets to show the efficacy of the proposed approach.

Improving visual noise insensitivity in small vocabulary audio visual speech recognition applications

Visual noise insensitivity is important to audio visual speech recognition (AVSR). Visual noise can take on a number of forms such as varying frame rate, occlusion, lighting or speaker variabilities. The use of a high dimensional secondary classifier on the word likelihood scores from both the audio and video modalities is investigated for the purposes of adaptive fusion. Preliminary results are presented demonstrating performance above the catastrophic fusion boundary for our confidence measure irrespective of the type of visual noise presented to it. Our experiments were restricted to small vocabulary applications.

A survey on compressed domain video analysis techniques

Image and video analysis requires rich features that can characterize various aspects of visual information. These rich features are typically extracted from the pixel values of the images and videos, which require huge amount of computation and seldom useful for real-time analysis. On the contrary, the compressed domain analysis offers relevant information pertaining to the visual content in the form of transform coefficients, motion vectors, quantization steps, coded block patterns with minimal computational burden. The quantum of work done in compressed domain is relatively much less compared to pixel domain. This paper aims to survey various video analysis efforts published during the last decade across the spectrum of video compression standards. In this survey, we have included only the analysis part, excluding the processing aspect of compressed domain. This analysis spans through various computer vision applications such as moving object segmentation, human action recognition, indexing, retrieval, face detection, video classification and object tracking in compressed videos.

Tracking Human Position and Lower Body Parts Using Kalman and Particle Filters Constrained by Human Biomechanics

In this paper, a novel framework for visual tracking of human body parts is introduced. The approach presented demonstrates the feasibility of recovering human poses with data from a single uncalibrated camera by using a limb-tracking system based on a 2-D articulated model and a double-tracking strategy. Its key contribution is that the 2-D model is only constrained by biomechanical knowledge about human bipedal motion, instead of relying on constraints that are linked to a specific activity or camera view. These characteristics make our approach suitable for real visual surveillance applications. Experiments on a set of indoor and outdoor sequences demonstrate the effectiveness of our method on tracking human lower body parts. Moreover, a detail comparison with current tracking methods is presented.

Video Activism and the Ambiguities of Counter-Surveillance

This paper examines the use of visual technologies by political activists in protest situations to monitor police conduct. Using interview data with Australian video activists, this paper seeks to understand the motivations, techniques and outcomes of video activism, and its relationship to counter-surveillance and police accountability. Our data also indicated that there have been significant transformations in the organization and deployment of counter-surveillance methods since 2000, when there were large-scale protests against the World Economic Forum meeting in Melbourne accompanied by a coordinated campaign that sought to document police misconduct. The paper identifies and examines two inter-related aspects of this: the act of filming and the process of dissemination of this footage. It is noted that technological changes over the last decade have led to a proliferation of visual recording technologies, particularly mobile phone cameras, which have stimulated a corresponding proliferation of images. Analogous innovations in internet communications have stimulated a coterminous proliferation of potential outlets for images Video footage provides activists with a valuable tool for safety and publicity. Nevertheless, we argue, video activism can have unintended consequences, including exposure to legal risks and the amplification of official surveillance. Activists are also often unable to control the political effects of their footage or the purposes to which it is used. We conclude by assessing the impact that transformations in both protest organization and media technologies might have for counter-surveillance techniques based on visual surveillance.

Translation Invariance in Object Recognition, and Its Relation to Other Visual Transformations

Human object recognition is generally considered to tolerate changes of the stimulus position in the visual field. A number of recent studies, however, have cast doubt on the completeness of translation invariance. In a new series of experiments we tried to investigate whether positional specificity of short-term memory is a general property of visual perception. We tested same/different discrimination of computer graphics models that were displayed at the same or at different locations of the visual field, and found complete translation invariance, regardless of the similarity of the animals and irrespective of direction and size of the displacement (Exp. 1 and 2). Decisions were strongly biased towards same decisions if stimuli appeared at a constant location, while after translation subjects displayed a tendency towards different decisions. Even if the spatial order of animal limbs was randomized ("scrambled animals"), no deteriorating effect of shifts in the field of view could be detected (Exp. 3). However, if the influence of single features was reduced (Exp. 4 and 5) small but significant effects of translation could be obtained. Under conditions that do not reveal an influence of translation, rotation in depth strongly interferes with recognition (Exp. 6). Changes of stimulus size did not reduce performance (Exp. 7). Tolerance to these object transformations seems to rely on different brain mechanisms, with translation and scale invariance being achieved in principle, while rotation invariance is not.

Communication mechanisms and middleware for distributed video surveillance

A new generation of advanced surveillance systems is being conceived as a collection of multi-sensor components such as video, audio and mobile robots interacting in a cooperating manner to enhance situation awareness capabilities to assist surveillance personnel. The prominent issues that these systems face are: the improvement of existing intelligent video surveillance systems, the inclusion of wireless networks, the use of low power sensors, the design architecture, the communication between different components, the fusion of data emerging from different type of sensors, the location of personnel (providers and consumers) and the scalability of the system. This paper focuses on the aspects pertaining to real-time distributed architecture and scalability. For example, to meet real-time requirements, these systems need to process data streams in concurrent environments, designed by taking into account scheduling and synchronisation. The paper proposes a framework for the design of visual surveillance systems based on components derived from the principles of Real Time Networks/Data Oriented Requirements Implementation Scheme (RTN/DORIS). It also proposes the implementation of these components using the well-known middleware technology Common Object Request Broker Architecture (CORBA). Results using this architecture for video surveillance are presented through an implemented prototype.

An eye for an eye: robust motion segmentation by applying the latest in human vision research

Automatically extracting interesting objects from videos is a very challenging task and is applicable to many research areas such robotics, medical imaging, content based indexing and visual surveillance. Automated visual surveillance is a major research area in computational vision and a commonly applied technique in an attempt to extract objects of interest is that of motion segmentation. Motion segmentation relies on the temporal changes that occur in video sequences to detect objects, but as a technique it presents many challenges that researchers have yet to surmount. Changes in real-time video sequences not only include interesting objects, environmental conditions such as wind, cloud cover, rain and snow may be present, in addition to rapid lighting changes, poor footage quality, moving shadows and reflections. The list provides only a sample of the challenges present. This thesis explores the use of motion segmentation as part of a computational vision system and provides solutions for a practical, generic approach with robust performance, using current neuro-biological, physiological and psychological research in primate vision as inspiration.