Ciril Bohak

Kinect Web Kiosk Framework

In this paper we present a web kiosk framework based on Kinect sensor. The main idea is to use the framework for creation of simple interactive presentations for informing, advertising and presenting knowledge to the public. The use of such a framework simplifies adaptation of existing web materials for presentation with the kiosk. We can also make use of touchless interaction for browsing through the interactive content, to animate the user and encourage her to spend more time browsing the presented content. We present the structure of the framework and a simple case study on using the framework as an interactive presentation platform and as an education resource. The developed framework has been used for presenting information on educational programs at Faculty of Computer and Information Science, University of Ljubljana.

Probabilistic Segmentation of Folk Music Recordings

The paper presents a novel method for automatic segmentation of folk music field recordings. The method is based on a distance measure that uses dynamic time warping to cope with tempo variations and a dynamic programming approach to handle pitch drifting for finding similarities and estimating the length of repeating segment. A probabilistic framework based on HMM is used to find segment boundaries, searching for optimal match between the expected segment length, between-segment similarities, and likely locations of segment beginnings. Evaluation of several current state-of-the-art approaches for segmentation of commercial music is presented and their weaknesses when dealing with folk music are exposed, such as intolerance to pitch drift and variable tempo. The proposed method is evaluated and its performance analyzed on a collection of 206 folk songs of different ensemble types: solo, two- and three-voiced, choir, instrumental, and instrumental with singing. It outperforms current commercial music segmentation methods for noninstrumental music and is on a par with the best for instrumental recordings. The method is also comparable to a more specialized method for segmentation of solo singing folk music recordings.

Ladar data generation fused with virtual targets and visualization for small drone detection system

For detection of a small target using electro-optical systems, multi-band 2D image sensors are used such as visible, NIR, MWIR, and LWIR. However, 2D imaging systems are not capable to detect a very small target and they are also not capable of calculating target 3D position coordinates to develop the strategic counter method. 3D sensors (e.g. Lidar, RGBD and stereo camera) are utilized to control unmanned vehicles for detecting threats and response for specific situations. Conventional Lidar systems are unable to detect small drone threat at distances higher than their maximum detecting range of 100 ∼ 120 meters. To overcome this limitation, laser radar (LADAR) systems are being developed, which allow the detection at distances up to 2 kilometers. In the development of LADAR, it is difficult to acquire datasets that contain cases of long distant targets. In this study, a fusion data generation with virtual targets technique based on minimum real LADAR initial map dataset is proposed, and precise small target detection method using voxel-based clustering and classification are studied. We present the process of data fusion generation and the experimental results for a small target detection. The presented approach also includes effective visualization of high-resolution 3D data and the results of small target detection in real time. This study is expected to contribute to the optimization of a drone threat detection system for various environments and characteristics.

Real-time interactive platform-agnostic volumetric path tracing in webGL 2.0

Path tracing has become a de facto standard for photo-realistic rendering due to its conceptual and algorithmic simplicity. Over the last few years, it has been successfully applied to the rendering of participating media, although it has not seen widespread adoption. Most implementations are targeted at specific platforms or hardware, which makes them difficult to deploy or extend. However, recent advancements in web technologies enable us to access graphics hardware from a web browser in a platform-agnostic manner. Therefore, in this paper, we present an implementation of a state-of-the-art volumetric path tracer developed in JavaScript using WebGL 2.0. The presented solution supports the use of arbitrary 2D transfer functions and heterogeneous volumetric data, aims to be interactive, platform-agnostic, easily extensible, and runs in real-time - both on desktop and mobile devices.

Vulkan Abstraction Layer for Large Data Remote Rendering System

New graphics APIs require users to implement a lot of needed functionality, such as memory management, by themselves. In this paper we present an abstraction layer build on top of such API, in our case the Vulkan API, for purpose of off-screen rendering of large data. We also present a use case for such abstraction layer implementation – a remote rendering system for simple Path Tracing accessible through web-based client-side application. The preliminary evaluation results show that implementation of simple Path Tracer is significantly faster then comparable implementation in OpenCL. In conclusion we also present possible extension and improvements of the developed abstraction layer.

Web-Based Vascular Flow Simulation Visualization with Lossy Data Compression for Fast Transmission.

In this paper, we present a web-based system for visualization of flow simulation results in the vascular system for use with consumer-level hardware. The presented tool allows users to design, execute and visualize a flow simulation with a simple workflow on a desktop computer or a mobile device. The web interface allows users to select a vascular model, define the flow simulation parameters, execute the simulation, and interactively visualize the simulation results in real time using multiple visualization techniques. The server-side prepares the model for simulation and performs the simulation using SimVascular. To provide a more efficient transfer of the large amounts of simulation results to the web client, as well as reduce storage requirements on the server, we introduce a novel hybrid lossy compression method. The method uses an octree data subdivision approach combined with an iterative approach that regresses the data points to a B-Spline volume. The evaluation results show that our method achieves compression ratios of up to 5.7 for the tested examples at a given error rate, comparable to other approaches while specifically intended for visualization purposes.

Web-Based Real-Time LADAR Data Visualization with Multi-user Collaboration Support

In this paper we present a web-based visualization system developed for visualizing real-time point cloud data obtained from LADAR (or other) sensors. The system allows direct visualization of captured data, visualization of data from database or visualization of preprocessed data (e.g. labeled or classified data). The system allows the concurrent visualization from same or different data-sources on multiple clients in the web browser. Due to the use of modern web technologies the client can also be used on mobile devices. The system is developed using modern client- and server-side web technologies. The system allows connection with an existing LADAR sensor grabber applications through use of UDP sockets. Both server- and client-side parts of the system are modular and allow the integration of newly developed modules and designing a specific work-flow scenarios for target end-user groups. The system allows the interactive visualization of datasets with millions of points as well as streaming visualization with high throughput speeds.