Catálogo de publicaciones - libros

A movie captured by a wearable camera affixed to an actor’s body gives audiences the sense of “immerse in the movie”. The raw movie captured by wearable camera needs stabilization with jitters due to ego-motion. However, conventional approaches often fail in accurate ego-motion estimation when there are moving objects in the image and no sufficient feature pairs provided by background region. To address this problem, we proposed a new approach that utilizes an additional synchronized video captured by the camera attached on the foreground object (another actor). Formally we configure above sensor system as two face-to-face moving cameras. Then we derived the relations between four views including two consecutive views from each camera. The proposed solution has two steps. Firstly we calibrate the extrinsic relationship of two cameras with an AX=XB formulation, and secondly estimate the motion using calibration matrix. Experiments verify that this approach can recover from failures of conventional approach and provide acceptable stabilization results for real data.

We propose a statistical motion model for sequential Bayesian tracking, called the , and show an adaptive particle filter algorithm with the motion model. It predicts the current state with the help of optical flows, i.e., it explores the state space with information based on the current and previous images of an image sequence. In addition, we introduce an automatic method for adjusting the variance of the motion model, which parameter is manually determined in most particle filters. In experiments with synthetic and real image sequences, we compare the proposed motion model with a random walk model, which is a widely used model for tracking, and show the proposed model outperform the random walk model in terms of accuracy even though their execution times are almost the same.

In this paper, we brought out a noise-insensitive pixel-wise object tracking algorithm whose kernel is a new reliable data grouping algorithm that introduces the reliability evaluation into the existing K-means clustering (called as RK-means clustering). The RK-means clustering concentrates on two problems of the existing K-mean clustering algorithm: 1) the unreliable clustering result when the noise data exists; 2) the bad/wrong clustering result caused by the incorrectly assumed number of clusters. The first problem is solved by evaluating the reliability of classifying an unknown data vector according to the triangular relationship among it and its two nearest cluster centers. Noise data will be ignored by being assigned low reliability. The second problem is solved by introducing a new group merging method that can delete pairs of ”too near” data groups by checking their variance and average reliability, and then combining them together. We developed a video-rate object tracking system (called as RK-means tracker) with the proposed algorithm. The extensive experiments of tracking various objects in cluttered environments confirmed its effectiveness and advantages.

We present a new approach towards efficient and robust tracking by incorporating the efficiency of the mean shift algorithm with the robustness of the particle filtering. The mean shift tracking algorithm is robust and effective when the representation of a target is sufficiently discriminative, the target does not jump beyond the bandwidth, and no serious distractions exist. In case of sudden motion, the particle filtering outperforms the mean shift algorithm at the expense of using a large particle set. In our approach, the mean shift algorithm is used as long as it provides reasonable performance. Auxiliary particles are introduced to conquer the distraction and sudden motion problems when such threats are detected. Moreover, discriminative features are selected according to the separation of the foreground and background distributions. We demonstrate the performance of our approach by comparing it with other trackers on challenging image sequences.

This paper presents an efficient indexing and retrieval scheme for searching in document image databases. In many non-European languages, optical character recognizers are not very accurate. Word spotting - word image matching - may instead be used to retrieve word images in response to a word image query. The approaches used for word spotting so far, dynamic time warping and/or nearest neighbor search, tend to be slow. Here indexing is done using locality sensitive hashing (LSH) - a technique which computes multiple hashes - using word image features computed at word level. Efficiency and scalability is achieved by content-sensitive hashing implemented through approximate nearest neighbor computation. We demonstrate that the technique achieves high precision and recall (in the 90% range), using a large image corpus consisting of seven Kalidasa’s (a well known Indian poet of antiquity) books in the Telugu language. The accuracy is comparable to using dynamic time warping and nearest neighbor search while the speed is orders of magnitude better - 20000 word images can be searched in milliseconds.

This paper proposes a novel method of estimating 3-D hand posture from images observed in complex backgrounds. Conventional methods often cause mistakes by mis-matches of local image features. Our method considers possibility of the mis-match between each posture model appearance and the other model appearances in a Baysian stochastic estimation form by introducing a novel likelihood concept “Mistakenly Matching Likelihood (MML)“. The correct posture model is discriminated from mis-matches by MML-based posture candidate evaluation. The method is applied to hand tracking problem in complex backgrounds and its effectiveness is shown.

We consider the problem of monocular 3d body pose tracking from video sequences. This task is inherently ambiguous. We propose to learn a generative model of the relationship of body pose and image appearance using a sparse kernel regressor. Within a particle filtering framework, the potentially multimodal posterior probability distributions can then be inferred. The 2d bounding box location of the person in the image is estimated along with its body pose. Body poses are modelled on a low-dimensional manifold, obtained by LLE dimensionality reduction. In addition to the appearance model, we learn a prior model of likely body poses and a nonlinear dynamical model, making both pose and bounding box estimation more robust. The approach is evaluated on a number of challenging video sequences, showing the ability of the approach to deal with low-resolution images and noise.

This paper proposes a novel volume-based motion capture method using a bottom-up analysis of volume data and an example topology database of the human body. By using a two-step graph matching algorithm with many example topological graphs corresponding to postures that a human body can take, the proposed method does not require any initial parameters or iterative convergence processes, and it can solve the changing topology problem of the human body. First, three-dimensional curved lines (skeleton) are extracted from the captured volume data using the thinning process. The skeleton is then converted into an attributed graph. By using a graph matching algorithm with a large amount of example data, we can identify the body parts from each curved line in the skeleton. The proposed method is evaluated using several video sequences of a single person and multiple people, and we can confirm the validity of our approach.

A logical dynamic programming (DP) matching algorithm is proposed for extracting similar subpatterns from two sequential patterns. In the proposed algorithm, local similarity between two patterns is measured by a logical function, called support. The DP matching with the support can extract all similar subpatterns simultaneously while compensating nonlinear fluctuation. The performance of the proposed algorithm was evaluated qualitatively and quantitatively via an experiment of extracting motion primitives, i.e., common subpatterns in gesture patterns of different classes.

In this paper we propose an efficient normalized cross correlation (NCC) algorithm for pattern matching based on adaptive multilevel successive elimination. This successive elimination scheme is applied in conjunction with an upper bound for the cross correlation derived from Cauchy-Schwarz inequality. To apply the successive elimination, we partition the summation of cross correlation into different levels with the partition order determined by the gradient energies of the partitioned regions in the template. Thus, this adaptive multi-level successive elimination scheme can be employed to early reject most candidates to reduce the computational cost. Experimental results show the proposed algorithm is very efficient for pattern matching under different lighting conditions.