Accurate computation of image motion enables the enhancement of image sequences. In scenes having multiple moving objects the motion computation is performed together with object segmentation by using a unique temporal integration approach. After computing the motion for the di erent image regions, these regions can be enhanced by fusing several successive frames covering the same region. Enhan...

An adaptive object recognition scheme for image sequences of many object scenes is described. The scheme is applied for t r d c object recognition under ego-motion. The recursive estimation of object states is performed by an extended Kalman Filter with modified error estimation, which is a neural network learning process. This new feature allows to separate the judgment needed for selection of...

This survey focuses on some of the most important and recent algorithms for pedestrian detection in video sequences taken from a camera mounted on a car. The longterm goal of this line of research is to develop vision-based systems that could be successfully implemented in cars for driver support in urban scenes or for autonomous driving. Some of the surveyed systems have already proved their e...

Estimating motion in scenes containing multiple motions remains a di cult problem for computer vision. Here we describe a novel recurrent network architecture which solves this problem by simultaneously estimating motion and segmenting the scene. The network is comprised of locally connected units which carry out simple calculations in parallel. We present simulation results illustrating the su...

Extracting in-play scenes in sport videos is essential for quantitative analysis and effective video browsing of the sport activities. Game analysis of badminton as of the other racket sports requires detecting the start and end of each rally period in an automated manner. This paper describes an automatic serve scene detection method employing cubic higher-order local auto-correlation (CHLAC) ...

Formal models of geographic space should support reasoning about its static and dynamic properties, its objects, their behaviors, and the relationships between them. Image schemas, used to embody spatiotemporal experiential abstractions, capture high-level perceptual concepts but do not have generally accepted formalizations. This paper provides a method for formally representing topological an...

Image-Based Virtual Environments are alternative approaches for virtual reality applications to generate real world scenes in virtual environments with low computation cost. This paper presents an approach to creating interactive image-based virtual reality environments in stereoscopic modes using binocular stereoscopic omnidirectional image sequences. The Omnidirectional Binocular Stereoscopic...

This paper attempts to evaluate quantitatively the performance trade-offs involved in applying object-oriented software methodologies to region-based representations of images and image sequences. Spatial or temporal multiscale representations of these regions are used as an example for the different levels of granularity that can be encapsulated separately as self-contained objects within thes...

Cardiodynamic and hemodynamic reactions to emotion-eliciting film sequences were investigated. Thirty-two healthy subjects (12 women, 20 men) were randomly assigned to one of two groups. In the first group, anger was induced using selected scenes of the film "Ragtime." In the second group, scenes of the film "The Shining" were chosen to elicit fear. A documentary film was used as a baseline sti...

Accurate dense 3D reconstruction of dynamic scenes from natural images is still very challenging. Most previous methods rely on a large number of fixed cameras to obtain good results. Some of these methods further require separation of static and dynamic points, which are usually restricted to scenes with known background. We propose a novel dense depth estimation method which can automatically...