Michal Lewandowski

Temporal Extension of Laplacian Eigenmaps for Unsupervised Dimensionality Reduction of Time Series

A novel non-linear dimensionality reduction method, called Temporal Laplacian Eigenmaps, is introduced to process efficiently time series data. In this embedded-based approach, temporal information is intrinsic to the objective function, which produces description of low dimensional spaces with time coherence between data points. Since the proposed scheme also includes bidirectional mapping between data and embedded spaces and automatic tuning of key parameters, it offers the same benefits as mapping-based approaches. Experiments on a couple of computer vision applications demonstrate the superiority of the new approach to other dimensionality reduction method in term of accuracy. Moreover, its lower computational cost and generalisation abilities suggest it is scalable to larger datasets.

View and style-independent action manifolds for human activity recognition

We introduce a novel approach to automatically learn intuitive and compact descriptors of human body motions for activity recognition. Each action descriptor is produced, first, by applying Temporal Laplacian Eigenmaps to view-dependent videos in order to produce a stylistic invariant embedded manifold for each view separately. Then, all view-dependent manifolds are automatically combined to discover a unified representation which model in a single three dimensional space an action independently from style and viewpoint. In addition, a bidirectional nonlinear mapping function is incorporated to allow projecting actions between original and embedded spaces. The proposed framework is evaluated on a real and challenging dataset (IXMAS), which is composed of a variety of actions seen from arbitrary viewpoints. Experimental results demonstrate robustness against style and view variation and match the most accurate action recognition method.

Structural Laplacian Eigenmaps for Modeling Sets of Multivariate Sequences

A novel embedding-based dimensionality reduction approach, called structural Laplacian Eigenmaps, is proposed to learn models representing any concept that can be defined by a set of multivariate sequences. This approach relies on the expression of the intrinsic structure of the multivariate sequences in the form of structural constraints, which are imposed on dimensionality reduction process to generate a compact and data-driven manifold in a low dimensional space. This manifold is a mathematical representation of the intrinsic nature of the concept of interest regardless of the stylistic variability found in its instances. In addition, this approach is extended to model jointly several related concepts within a unified representation creating a continuous space between concept manifolds. Since a generated manifold encodes the unique characteristic of the concept of interest, it can be employed for classification of unknown instances of concepts. Exhaustive experimental evaluation on different datasets confirms the superiority of the proposed methodology to other state-of-the-art dimensionality reduction methods. Finally, the practical value of this novel dimensionality reduction method is demonstrated in three challenging computer vision applications, i.e., view-dependent and view-independent action recognition as well as human-human interaction classification.

Are current monocular computer vision systems for human action recognition suitable for visual surveillance applications

Since video recording devices have become ubiquitous, the automated analysis of human activity from a single uncalibrated video has become an essential area of research in visual surveillance. Despite variability in terms of human appearance and motion styles, in the last couple of years, a few computer vision systems have reported very encouraging results. Would these methods be already suitable for visual surveillance applications? Alas, few of them have been evaluated in the two most challenging scenarios for an action recognition system: view independence and human interactions. Here, first a review of monocular human action recognition methods that could be suitable for visual surveillance is presented. Then, the most promising frameworks, i.e. methods based on advanced dimensionality reduction, bag of words and random forest, are described and evaluated on IXMAS and UT-Interaction datasets. Finally, suitability of these systems for visual surveillance applications is discussed.

Re-identification of pedestrians with variable occlusion and scale

This paper presents results from experiments designed to measure the accuracy with which people can be reidentified using multiple visual surveillance observations. Two public data sets are used: VIPeR and a new public data set, V-47. The re-identification method is a Large Margin Nearest Neighbour classifier using feature vectors constructed from overlapping block histograms. The experiments investigate the performance with respect to the level of occlusion, the training regime, specificity of the domain and the resolution of the observations. A method is proposed that reduces the adverse impact of occlusions, when present; and increases the beneficial impact of higher resolution data, when available.

Generalized Laplacian eigenmaps for modeling and tracking human motions.

This paper presents generalized Laplacian eigenmaps, a novel dimensionality reduction approach designed to address stylistic variations in time series. It generates compact and coherent continuous spaces whose geometry is data-driven. This paper also introduces graph-based particle filter, a novel methodology conceived for efficient tracking in low dimensional space derived from a spectral dimensionality reduction method. Its strengths are a propagation scheme, which facilitates the prediction in time and style, and a noise model coherent with the manifold, which prevents divergence, and increases robustness. Experiments show that a combination of both techniques achieves state-of-the-art performance for human pose tracking in underconstrained scenarios.

Automatic configuration of spectral dimensionality reduction methods

We propose an advanced framework for the automatic configuration of spectral dimensionality reduction methods. This is achieved by introducing, first, the mutual information measure to assess the quality of discovered embedded spaces. Secondly, unsupervised Radial Basis Function network is designated for mapping between spaces where the learning process is derived from graph theory and based on Markov cluster algorithm. Experiments on synthetic and real datasets demonstrate the effectiveness of the proposed methodology.

Human pose tracking in low dimensional space enhanced by limb correction

This paper proposes a two-level 3D human pose tracking method for a specific action captured by several cameras. The generation of pose estimates relies on fitting a 3D articulated model on a Visual Hull generated from the input images. First, an initial pose estimate is constrained by a low dimensional manifold learnt by Temporal Laplacian Eigenmaps. Then, an improved global pose is calculated by refining individual limb poses. The validation of our method uses a public standard dataset and demonstrates its accurate and computational efficiency.

Automatic configuration of spectral dimensionality reduction methods for 3D human pose estimation

In this paper, our main contribution is a framework for the automatic configuration of any spectral dimensionality reduction methods. This is achieved, first, by introducing the mutual information measure to assess the quality of discovered embedded spaces. Secondly, we overcome the deficiency of mapping function in spectral dimensionality reduction approaches by proposing data projection between spaces based on fully automatic and dynamically adjustable Radial Basis Function network. Finally, this automatic framework is evaluated in the context of 3D human pose estimation. We demonstrate mutual information measure outperforms all current space assessment metrics. Moreover, experiments show the mapping associated to the induced embedded space displays good generalization properties. In particular, it allows improvement of accuracy by around 30% when refining 3D pose estimates of a walking sequence produced by an activity independent method.

Tracklet Reidentification in Crowded Scenes Using Bag of Spatio-temporal Histograms of Oriented Gradients

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% of the first rank tracklet associations.