Stephan Kopf

COMPASS: A probabilistic indoor positioning system based on 802.11 and digital compasses

Positioning systems are one of the key elements required by location-based services. This paper presents the design, implementation and analysis of a positioning system called COMPASS which is based on 802.11-compliant network infrastructure and digital compasses. On the mobile device, COMPASS samples the signal strength values of different access points in its communication range and utilizes the orientation of the user to preselect a subset of the training data. The remaining training data is used by a probabilistic positioning algorithm to determine the position of the user. While prior systems show limited accuracy due to blocking effects caused by the human body, we apply digital compasses to detect the orientations of the users so that we can deal with these blocking effects. After a short period of training our COMPASS system achieves an average error distance of less than 1.65 meters in our experimental environment of 312 square meters.

FSCAV: fast seam carving for size adaptation of videos

The presentation of multimedia data and especially of high resolution videos on small mobile devices is still a great challenge today. Both cropping of borders and scaling of frames may result in the removal of essential content of videos or lost details due to the reduced size of the visual content. Another major problem emerges if the aspect ratio of the original video and the display of the mobile device differ. User evaluations indicate that changing the aspect ratio may reduce the visual quality of videos significantly. In this paper, we present the new FSCAV algorithm (Fast Seam Carving for Size Adaptation of Videos) to adapt the size of videos to the limited display resolution and different aspect ratios of handheld mobile devices. The general idea of the seam carving algorithm for still images is to remove seams in images so that the essential content is preserved. We extended this technique which works very well for images to create videos without jitter or visible artifacts. A major feature of our FSCAV algorithm is the low computational complexity which enables an efficient adaptation of videos to small screens. Nevertheless, severe distortions are clearly visible in some shots of the adapted videos. We present a new heuristic to identify shots with such a low visual quality. If the quality drops below a threshold, a different adaptation technique is used for this shot (e.g., scaling or cropping). User evaluations confirm a very high visual quality of our approach.

An automatic cameraman in a lecture recording system

We describe the design and implementation of an automatic cameraman for lecture recording. A major problem with traditional lecture recordings is that they tend to be boring for the students, especially if only the slides and the audio of the lecturer are pre-sented. In a first step, we determine the tasks a real cameraman would have, in particular with respect to liveliness of the video. We then adapt these tasks to a computer system and show in detail how they can be implemented. In a second step, we describe how our algorithms support the virtual director system into which the automatic cameraman is integrated. We conclude that lecture re-cordings can be much more lively and interesting using our approach.

Algorithms for video retargeting

The visualization of high resolution video on small mobile devices is still a great challenge today. Most critical are the limited display resolution and different aspect ratios of handheld mobile devices. So far, there is no retargeting algorithm available that guarantees good results for all videos. We introduce a new video retargeting approach that reduces the resolution while preserving as much of the relevant content as possible. A central component of the system selects the most suitable algorithm to adapt a given shot. We have implemented two retargeting algorithms: a region of interest (ROI) based technique, and a fast implementation of seam carving for size adaptation of videos (FSCAV). The ROI-based retargeting detects important regions like faces, objects, text, and contrast-based saliency regions. A rectangular window within the larger frame is selected that defines the visible area of the target video. If several relevant regions are detected, an artificial camera motion (pan, tilt, or zoom) may change the selected view within a shot. For seam carving, we present two extensions: The first reduces the distortion of straight lines (lines may become curved or disconnected); the second avoids jitter in the target video, limits the large memory requirements and computational effort of seam carving, and makes it applicable to video retargeting. In addition, we present a heuristic that estimates the visual quality of the target video. If the quality drops below a threshold, the ROI-based retargeting is used for this shot. User evaluations confirm a very high visual quality of our approach.

Seam Carving with Improved Edge Preservation

In this paper, we propose a new method to adapt the resolution of images to the limited display resolution of mobile devices. We use the seam carving technique to identify and remove less relevant content in images. Seam carving achieves a high adaptation quality for landscape images and distortions caused by the removal of seams are very low compared to other techniques like scaling or cropping. However, if an image depicts objects with straight lines or regular patterns like buildings, the visual quality of the adapted images is much lower. Errors caused by seam carving are especially obvious if straight lines become curved or disconnected. In order to preserve straight lines, our algorithm applies line detection in addition to the normal energy function of seam carving. The energy in the local neighborhood of the intersection point of a seam and a straight line is increased to prevent other seams from removing adjacent pixels. We evaluate our improved seam carving algorithm and compare the results with regular seam carving. In case of landscape images with no straight lines, traditional seam carving and our enhanced approach lead to very similar results. However, in the case of objects with straight lines, the quality of our results is significantly better.

Enhancing curvature scale space features for robust shape classification

The curvature scale space (CSS) technique, which is also part of the MPEG-7 standard, is a robust method to describe complex shapes. The central idea is to analyze the curvature of a shape and derive features from inflection points. A major drawback of the CSS method is its poor representation of convex segments: Convex objects cannot be represented at all due to missing inflection points. We have extended the CSS approach to generate feature points for concave and convex segments of a shape. This generic approach is applicable to arbitrary objects. In the experimental results, we evaluate as a comprehensive example the automatic recognition of characters in images and videos.

A Virtual Camera Team for Lecture Recording

We present the design and implementation of a virtual camera team for recording classroom lectures. Our approach to lecture recording, with well-defined tasks for each module, has two significant advantages. First, the workload is distributed; for example, the camera-operator modules and not the director module produce the images. Second, its easier to implement complex cinematographic rules using the well-defined roles of the virtual team members and the communication between them. In this way, the virtual camera teams behavior mimics the behavior of a human camera team and thus leads to more lively recordings than earlier approaches.

Automatic scaling and cropping of videos for devices with limited screen resolution

A large number of previously recorded videos cannot be directly visualized on mobile devices like PDAs or mobile phones due to an inappropriate screen resolution of their displays. Transcoding can be used to change the resolution, however, the usual transcoding algorithms have problems preserving the semantic content. For instance, superimposed text is unreadable if the character size drops below a certain value. In this paper, we present a novel adaptation algorithm to scale and crop videos while preserving their semantic content. Semantic features in a shot are combined to select a suitable region to be resented in the adapted video.

Optimal shutter speed sequences for real-time HDR video

A technique to create High Dynamic Range (HDR) video frames is to capture Low Dynamic Range (LDR) images at varying shutter speeds. They are then merged into a single image covering the entire brightness range of the scene. While shutter speeds are often chosen to vary by a constant factor, we propose an adaptive approach. The scenes histogram together with functions judging the contribution of an LDR exposure to the HDR result are used to compute a sequence of shutter speeds. This sequence allows for the estimation of the scenes radiance map with a high degree of accuracy. We show that, in comparison to the traditional approach, our algorithm achieves a higher quality of the HDR image for the same number of captured LDR exposures. Our algorithm is suited for creating HDR videos of scenes with varying brightness conditions in real-time, which applications like video surveillance benefit from.

Overhearing the Wireless Interface for 802.11-Based Positioning Systems

Not only the communication capabilities of 802.11, but also the capability to determine the position of mobile devices make 802.11 highly appealing for many application areas. Typically, a mobile device that wants to identify its position regularly performs active or passive scans to obtain the signal strength measurements of neighboring access points. However, so far, no investigations are known to have been launched into how regular scanning affects concurrent data transmissions from an end-user point of view. In this paper, we explore how common data communication is affected while actively or passively scanning at the same time. Furthermore, we present a novel scan scheme called monitor sniffing. Monitor sniffing exploits the fact that 802.11 operates on overlapping channels by overhearing the wireless interface. We have implemented our monitor sniffing algorithm using commodity 802.11g hardware, and we demonstrate that it does not disturb concurrent data communication