Dimitrios I. Kosmopoulos

Violence content classification using audio features

This work studies the problem of violence detection in audio data, which can be used for automated content rating. We employ some popular frame-level audio features both from the time and frequency domain. Afterwards, several statistics of the calculated feature sequences are fed as input to a Support Vector Machine classifier, which decides about the segment content with respect to violence. The presented experimental results verify the validity of the approach and exhibit a better performance than the other known approaches.

Robust Sequential Data Modeling Using an Outlier Tolerant Hidden Markov Model

Hidden Markov (chain) models using finite Gaussian mixture models as their hidden state distributions have been successfully applied in sequential data modeling and classification applications. Nevertheless, Gaussian mixture models are well known to be highly intolerant to the presence of untypical data within the fitting data sets used for their estimation. Finite Students t-mixture models have recently emerged as a heavier-tailed, robust alternative to Gaussian mixture models, overcoming these hurdles. To exploit these merits of Students t-mixture models in the context of a sequential data modeling setting, we introduce, in this paper, a novel hidden Markov model where the hidden state distributions are considered to be finite mixtures of multivariate Students t-densities. We derive an algorithm for the model parameters estimation under a maximum likelihood framework, assuming full, diagonal, and factor-analyzed covariance matrices. The advantages of the proposed model over conventional approaches are experimentally demonstrated through a series of sequential data modeling applications.

Multiclass defect detection and classification in weld radiographic images using geometric and texture features

In this paper, a method for the detection and classification of defects in weld radiographs is presented. The method has been applied for detecting and discriminating discontinuities in the weld images that may correspond to false alarms or defects such as worm holes, porosity, linear slag inclusion, gas pores, lack of fusion or crack. A set of 43 descriptors corresponding to texture measurements and geometrical features is extracted for each segmented object and given as input to a classifier. The classifier is trained to classify each of the objects it into one of the defect classes or characterize it as non-defect. Three fold cross validation was utilized and experimental results are reported for three different classifiers (Support Vector Machine, Neural Network, k-NN).

Audio-Visual fusion for detecting violent scenes in videos

In this paper we present our research towards the detection of violent scenes in movies, employing fusion methodologies, based on learning Towards this goal, a multi-step approach is followed: initially, automated auditory and visual processing and analysis is performed in order to estimate probabilistic measures regarding particular audio and visual related classes At a second stage, a meta-classification architecture is adopted, which combines the audio and visual information, in order to classify mid-term video segments as “violent” or “non-violent” The proposed scheme has been evaluated on a real dataset from 10 films.

Detecting abnormal human behaviour using multiple cameras

In this paper a bottom-up approach for human behaviour understanding is presented, using a multi-camera system. The proposed methodology, given a training set of normal data only, classifies behaviour as normal or abnormal, using two different criteria of human behaviour abnormality (short-term behaviour and trajectory of a person). Within this system an one-class support vector machine decides short-term behaviour abnormality, while we propose a methodology that lets a continuous Hidden Markov Model function as an one-class classifier for trajectories. Furthermore, an approximation algorithm, referring to the Forward Backward procedure of the continuous Hidden Markov Model, is proposed to overcome numerical stability problems in the calculation of probability of emission for very long observations. It is also shown that multiple cameras through homography estimation provide more precise position of the person, leading to more robust system performance. Experiments in an indoor environment without uniform background demonstrate the good performance of the system.

Automated inspection of gaps on the automobile production line through stereo vision and specular reflection

Abstract One of the most difficult tasks in the later stages of automobile assembly is the dimensional inspection of the gaps between the car body and the various panels fitted on it (doors, motor-hood, etc.). The employment of an automatic gap-measuring system would reduce the costs significantly and would offer high flexibility. However, this task is still performed by humans and only a few — still experimental — automatic systems have been reported. In this paper, we introduce a system for automated gap inspection that employs computer vision. It is capable of measuring the lateral and the range dimension of the gap (width and flush, correspondingly). The measurement installation consists of two calibrated stereo cameras and two infrared LED lamps, used for highlighting the edges of the gap through specular reflection. The gap is measured as the 3D distance between the highlighted edges. This method has significant advantages against the laser-based, gap-measuring systems, mainly due to its color independency. Our approach has been analytically described in 2D and extensively evaluated using synthetic as well as real gaps. The results obtained verify its robustness and its applicability in an industrial environment.

Hand Shape and 3D Pose Estimation Using Depth Data from a Single Cluttered Frame

This paper describes a method that, given an input image of a person signing a gesture in a cluttered scene, locates the gesturing arm, automatically detects and segments the hand and finally creates a ranked list of possible shape class, 3D pose orientation and full hand configuration parameters. The clutter-tolerant hand segmentation algorithm is based on depth data from a single image captured with a commercially available depth sensor, namely the Kinect TM . Shape and 3D pose estimation is formulated as an image database retrieval method where given a segmented hand the best matches are extracted from a large database of synthetically generated hand images. Contrary to previous approaches this clutter-tolerant method is all-together: user-independent, automatically detects and segments the hand from a single image (no multi-view or motion cues employed) and provides estimation not only for the 3D pose orientation but also for the full hand articulation parameters. The performance of this approach is quantitatively and qualitatively evaluated on a dataset of real images of American Sign Language (ASL) handshapes.

Signal Modeling and Classification Using a Robust Latent Space Model Based on

Factor analysis is a statistical covariance modeling technique based on the assumption of normally distributed data. A mixture of factor analyzers can be hence viewed as a special case of Gaussian (normal) mixture models providing a mathematically sound framework for attribute space dimensionality reduction. A significant shortcoming of mixtures of factor analyzers is the vulnerability of normal distributions to outliers. Recently, the replacement of normal distributions with the heavier-tailed Students-t distributions has been proposed as a way to mitigate these shortcomings and the treatment of the resulting model under an expectation-maximization (EM) algorithm framework has been conducted. In this paper, we develop a Bayesian approach to factor analysis modeling based on Students-t distributions. We derive a tractable variational inference algorithm for this model by expressing the Students-t distributed factor analyzers as a marginalization over additional latent variables. Our innovative approach provides an efficient and more robust alternative to EM-based methods, resolving their singularity and overfitting proneness problems, while allowing for the automatic determination of the optimal model size. We demonstrate the superiority of the proposed model over well-known covariance modeling techniques in a wide range of signal processing applications.

t

In this paper, we propose a novel online framework for behavior understanding, in visual workflows, capable of achieving high recognition rates in real-time. To effect online recognition, we propose a methodology that employs a Bayesian filter supported by hidden Markov models. We also introduce a novel re-adjustment framework of behavior recognition and classification by incorporating the users feedback into the learning process through two proposed schemes: a plain non-linear one and a more sophisticated recursive one. The proposed approach aims at dynamically correcting erroneous classification results to enhance the behavior modeling and therefore the overall classification rates. The performance is thoroughly evaluated under real-life complex visual behavior understanding scenarios in an industrial plant. The obtained results are compared and discussed.

Distributions

The monitoring of sleep patterns is of major importance for various reasons such as the detection and treatment of sleep disorders, the assessment of the effect of different medical conditions or medications on the sleep quality, and the assessment of mortality risks associated with sleeping patterns in adults and children. Sleep monitoring by itself is a difficult problem due to both privacy and technical considerations. The proposed system uses a combination of non-invasive sensors to assess and report sleep patterns: a contact-based pressure mattress and a non-contact 3D image acquisition device, which can complement each other. To evaluate our system, we used real data collected in Heracleia Lab’s assistive living apartment. Our system uses Machine Learning techniques to automatically analyze the collected data and recognize sleep patterns. It is non-invasive, as it does not disrupt the user’s usual sleeping behavior and it can be used both at the clinic and at home with minimal cost.