Daesik Jang

A real-time 3D workspace modeling with stereo camera

This paper presents a novel approach to real-time 3D modeling of workspace for manipulative robotic tasks. First, we establish the three fundamental principles that human uses for modeling and interacting with environment. These principles have led to the development of an integrated approach to real-time 3D modeling, as follows: 1) It starts with a rapid but approximate characterization of the geometric configuration of workspace by identifying global plane features. 2) It quickly recognizes known objects in workspace and replaces them by their models in database based on in-situ registration. 3) It models the geometric details on the fly adaptively to the need of the given task based on a multi-resolution octree representation. SIFT features with their 3D position data, referred to here as stereo-sis SIFT, are used extensively, together with point clouds, for fast extraction of global plane features, for fast recognition of objects, for fast registration of scenes, as well as for overcoming incomplete and noisy nature of point clouds. The experimental results show the feasibility of real-time and behavior-oriented 3D modeling of workspace for robotic manipulative tasks.

User oriented language model for face detection

This paper provides a novel approach for a user oriented language model for face detection. Even though there are many open source or commercial libraries to solve the problem of face detection, they are still hard to use because they require specific knowledge on details of algorithmic techniques. This paper proposes a high-level language model for face detection with which users can develop systems easily and even without specific knowledge on face detection theories and algorithms. Important conditions are firstly considered to categorize the large problem space of face detection. The conditions identified here are then represented as expressions in terms of a language model so that developers can use them to express various problems. Once the conditions are expressed by users, the proposed associated interpreter interprets the conditions to find and organize the best algorithms to solve the represented problem with corresponding conditions. We show a proof-of-concept implementation and some test and analyze example problems to show the ease of use and usability.

MediaDiver: viewing and annotating multi-view video

We propose to bring our novel rich media interface called MediaDiver demonstrating our new interaction techniques for viewing and annotating multiple view video. The demonstration allows attendees to experience novel moving target selection methods (called Hold and Chase), new multi-view selection techniques, automated quality of view analysis to switch viewpoints to follow targets, integrated annotation methods for viewing or authoring meta-content and advanced context sensitive transport and timeline functions. As users have become increasingly sophisticated when managing navigation and viewing of hyper-documents, they transfer their expectations to new media. Our proposal is a demonstration of the technology required to meet these expectations for video. Thus users will be able to directly click on objects in the video to link to more information or other video, easily change camera views and mark-up the video with their own content. The applications of this technology stretch from home video management to broadcast quality media production, which may be consumed on both desktop and mobile platforms.

Toward human-like real-time manipulation: From perception to motion planning

Human-like behavior is crucial for intelligent service robots that are to perform versatile tasks in day to day life. In this paper, an integrated approach to human-like manipulation is presented, which addresses realtime three-dimensional (3-D) workspace modeling and accessibility analysis for motion planning. The 3-D workspace modeling uses three main principles: identification of global geometric features, substitution of recognized known objects by corresponding solid models in the database and multi-resolution representation of unknown obstacles as required by the task at hand. Accessibility analysis is done through visibility tests. It complements and accelerates general motion planning. The experimental results demonstrate that the human-like behavior-oriented methods are sufficiently fast and robust to model 3-D workspace, and to plan and execute tasks for robotic manipulative applications.

Transforming cluster-based segmentation for use in OpenVL by mainstream developers

The majority of vision research focusses on advancing technical methods for image analysis, with a coupled increase in complexity and sophistication. The problem of providing access to these sophisticated techniques is largely ignored, leading to a lack of application by mainstream applications. We present a feature-based clustering segmentation algorithm with novel modifications to fit a developer-centred abstraction. This abstraction acts as an interface which accepts a description of segmentation in terms of properties (colour, intensity, texture, etc.), constraints (size, quantity) and priorities (biasing a segmentation). This paper discusses the modifications needed to fit the algorithm into the abstraction, which conditions of the abstraction it supports, and results of the various conditions demonstrating the coverage of the segmentation problem space. The algorithm modification process is discussed generally to help other researchers mould their algorithms to similar abstractions.

Recognition of 3D Objects from a Sequence of Images

The recognition of relatively big and rarely movable objects such as refrigerators and air conditioners, etc. is necessary because these objects can be crucial global features for Simultaneous Localization and Map building(SLAM) in indoor environment. In this paper, we propose a novel method to recognize these big objects using a sequence of 3D scenes. The particles which represent an object to be recognized are scattered into the 3D scene captured from an environment and then the probability of each particle is calculated by matching the 3D lines of the object model with them of the environment. Based on the probabilities and the degree of convergence of the particles, the object in the environment can be recognized and the position of the object can also be estimated. The experimental results show the feasibility of the suggested method based on particle filtering and its application to SLAM problems.