Catálogo de publicaciones - libros

Strong temporal correlation between adjacent frames of a video signal has been successfully exploited in standard video compression algorithms. In this work, we show that the temporal correlation in a video signal can also be used for fast video to reference image alignment. To this end, we first divide the input video sequence into groups of pictures (GOPs). Then for each GOP, only one frame is completely correlated with the reference image, while for the remaining frames, upper and lower bounds on the correlation coefficient () are calculated. These newly proposed bounds are significantly tighter than the existing Cauchy-Schwartz inequality based bounds on . These bounds are used to eliminate majority of the search locations and thus resulting in significant speedup, without effecting the value or location of the global maxima. In our experiments, up to 80% search locations are found to be eliminated and the speedup is up to five times the FFT based implementation and up to seven times the spatial domain techniques.

The ability to detect and recognize individuals is essential for an autonomous robot interacting with humans even if computational resources are usually rather limited. In general a small user group can be assumed for interaction. The robot has to distinguish between multiple users and further on between and persons. For solving this problem we propose an approach which integrates detection, recognition and tracking by formulating all tasks as binary classification problems. Because of its efficiency it is well suited for robots or other systems with limited resources but nevertheless demonstrates robustness and comparable results to state-of-the-art approaches. We use a common over-complete representation which is shared by the different modules. By means of the integral data structure an efficient feature computation is performed enabling the usage of this system for real-time applications such as for our autonomous robot .

More and more researchers focus their studies on multi-view activity recognition, because a fixed view could not provide enough information for recognition. In this paper, we use multi-view features to recognize six kinds of gymnastic activities. Firstly, shape-based features are extracted from two orthogonal cameras in the form of transform. Then a multi-view approach based on Fused HMM is proposed to combine different features for similar gymnastic activity recognition. Compared with other activity models, our method achieves better performance even in the case of frame loss.

We present an automated system for real-time marker-free motion capture from two calibrated webcams. For fast 3D shape and skin reconstructions, we extend Shape-From-Silhouette algorithms. The motion capture system is based on simple and fast heuristics to increase the efficiency. Multi-modal scheme using both shape and skin-parts analysis, temporal coherence, and human anthropometric constraints are adopted to increase the robustness. Thanks to fast algorithms, low-cost cameras and the fact that the system runs on a single computer, our system is perfectly suitable for home entertainment device. Results on real video sequences demonstrate our approach efficiency.

This paper describes a sophisticated method to track iris contour and to estimate eye gaze for blinking eyes with a monocular camera. A 3D eye-model that consists of eyeballs, iris contours and eyelids is designed that describes the geometrical properties and the movements of eyes. Both the iris contours and the eyelid contours are tracked by using this eye-model and a particle filter. This algorithm is able to detect “pure” iris contours because it can distinguish iris contours from eyelids contours. The eye gaze is described by the movement parameters of the 3D eye model, which are estimated by the particle filter during tracking. Other distinctive features of this algorithm are: 1) it does not require any special light sources (e.g. an infrared illuminator) and 2) it can operate at video rate. Through extensive experiments on real video sequences we confirmed the robustness and the effectiveness of our method.

This paper proposes a novel eye detection method using the MCT-based pattern correlation. The proposed method detects the face by the MCT-based AdaBoost face detector over the input image and then detects two eyes by the MCT-based AdaBoost eye detector over the eye regions. Sometimes, we have some incorrectly detected eyes due to the limited detection capability of the eye detector. To reduce the falsely detected eyes, we propose a novel eye verification method that employs the MCT-based pattern correlation map. We verify whether the detected eye patch is eye or non-eye depending on the existence of a noticeable peak. When one eye is correctly detected and the other eye is falsely detected, we can correct the falsely detected eye using the peak position of the correlation map of the correctly detected eye. Experimental results show that the eye detection rate of the proposed method is 98.7% and 98.8% on the Bern images and AR-564 images.

In the coming ubiquitous-computing society, an eyegaze interface will be one of the key technologies as an input device. Most of the conventional eyegaze tracking algorithms require specific light sources, equipments, devices, etc. In a previous work, the authors developed a simple eye-gaze detection system using a monocular video camera. This paper proposes a fast eye-gaze detection algorithm using the parametric template matching. In our algorithm, the iris extraction by the parametric template matching is applied to the eye-gaze detection based on physiological eyeball model. The parametric template matching can carry out an accurate sub-pixel matching by interpolating a few template images of a user’s eye captured in the calibration process for personal error. So, a fast calculation can be realized with keeping the detection accuracy. We construct an eye-gaze communication interface using the proposed algorithm, and verified the performance through key typing experiments using visual keyboard on display.

Smart camera is a camera that can not only see but also think and act. A smart camera is an embedded vision system which captures and processes image to extract application-specific information in real time. The brain of a smart camera is a special processing module that performs application specific information processing. The design of a smart camera as an embedded system is challenging because video processing has insatiable demand for performance and power, but at the same time embedded systems place considerable constraints on the design. We present our work to develop GestureCam, an FPGA-based smart camera built from scratch that can recognize simple hand gestures. The first completed version of GestureCam has shown promising real-time performance and is being tested in several desktop HCI (Human Computer Interface) applications.

This paper introduces a novel convex kernel based method for color constancy computation with explicit illuminant parameter estimation. A simple linear render model is adopted and the illuminants in a new scene that contains some of the color surfaces seen in the training image are sequentially estimated in a global optimization framework. The proposed method is fully data-driven and initialization invariant. Nonlinear color constancy can also be approximately solved in this kernel optimization framework with piecewise linear assumption. Extensive experiments on real-scene images validate the practical performance of our method.

Many real objects, such as faces, sculptures, or low-reliefs are composed of many detailed parts that can not be easily modeled by an artist nor by 3D scanning. In this paper, we propose a new shape from shading (SfS) approach to rapidly model details of these objects such as wrinkles and reliefs of surfaces from one photograph. The method first determines the surface’s flat areas in the photograph. Then, it constructs a graph of relative altitudes between each of these flat areas. We circumvent the ill-posed problem of shape from shading by having the user set if some of these flat areas are a local maximum or a local minimum; additional points can be added by the user ( at discontinuous creases) – this is the only user input. We use an intuitive mass-spring based minimization to determine the final position of these flat areas and a fast-marching method to generate the surface. This process can be iterated until the user is satisfied with the resulting surface. We illustrate our approach on real faces and low-relief photographs.