Gwang-Gook Lee

Automatic video summarizing tool using MPEG-7 descriptors for personal video recorder

We introduce an automatic video summarizing tool (AVST) for a personal video recorder. The tool utilizes MPEG-7 visual descriptors to generate a video index for a summary. The resulting index generates not only a preview of a movie but also allows non-linear access with thumbnails. In addition, the index supports the searching of shots similar to a desired one within saved video sequences. Moreover, simple shot-based video editing can readily be achieved using the generated index.

Automatic Estimation of Pedestrian Flow

Counting the amount of pedestrian flow is an important task for video surveillance applications. Most previous methods for pedestrian flow counting employed model-based detection or top-view cameras to measure pedestrian flow. However, those approaches are difficult to apply to realistic applications because of high complexity or specific camera set up. In this paper, a pixel count based method is proposed for pedestrian flow estimation. In the proposed method, image features such as foreground pixels and motion vectors are utilized as clues to find the number of people going through a gate. To estimate the number of pedestrians without any modeling or tracking, the number of foreground pixels is accumulated on the gate. Experiments on the PETS2006 dataset revealed that the proposed method can count the number of pedestrians successfully even for viewpoint changes.

A Method of Counting Pedestrians in Crowded Scenes

This paper proposes a method to automatically count the number of pedestrians in a video input of a crowed scene. The method proposed in this paper improves on our previous pedestrian counting method which estimates the number of pedestrians by accumulating low-level features (foreground pixels and motion vectors) on a virtual gate. To handle crowded scenes, the pedestrian counting process in this paper is weighted by the ratio of foreground pixels in the scene. The relationship between crowdedness and weighting factor is learned from 10,000 simulation images. Tests on real video sequences show that this method can successfully estimate the number of pedestrians with an accuracy of about 95%. Also, when compared to the previous method, the accuracy was increased by about 5% for highly crowded scenes. Moreover, the proposed method runs at an average rate of around 60 fps on a standard PC, which makes the algorithm realistic for multi-camera systems.

Pedestrian Counting Using an IR Line Laser

This paper presents a method of pedestrian counting using an IR line laser. In the proposed method, only reflection of IR line laser onto pedestrians is obtained using an IR filter. Hence acquisition of moving object images is extremely simple and reliable compared to other vision-based method. The number of pedestrians is counted using simple blob analysis and their moving directions are obtained from slopes of the blobs. Experiments showed that the proposed method is able to count the number of pedestrians effectively (with accuracy of about 95%) and efficiently (9 ms for one frame on a standard PC).

Highlight generation for basketball video using probabilistic excitement

With popularization of multimedia devices, semantic analysis of sports video has been widely studied. In this paper, we propose a highlight generation method for basketball games. To create a video highlight, the proposed method selects interesting shots by modeling excitements of the game using score information. For this purpose, a video is first segmented into shots and classified as play and nonplay shots. At the same time, score of the game is automatically extracted from video frames. To select interesting shots, which should be included to the highlight video, excitement of the game is estimated from the variation of game scores. Unlike previous event-based video analysis methods which focus on individual events, our proposed method is able to reflect the contents of a game by considering excitements. Also, because the excitement modeling uses score information only, it can easily be extended to other types of sports video which have scores.

Prototype Development of a Spatial Information Management System for Large-Scale Buildings

This paper presents a new model of a management system for large-scale buildings that aims to deal with both static and dynamic spatial information. For this purpose, 3D CAD, 3D GIS and image processing are integrated. The geometrical information of a building is managed by GIS using a database built from 3D CAD. Also, the dynamic spatial information of a building (i.e., flow size of pedestrians) is obtained using image processing. We implemented a prototype version of the proposed system to a real environment and it showed promising results for a running system covering a wide area.

A video summarization method for basketball game

There have been various research efforts on automatic summarization of sports video. However, most previous works were based on event detection and thus cannot reflect the semantic importance of scenes and content of a game. In this paper, a summarization method for basketball video is presented. The proposed method keeps track of score changes of the game by reading the numbers on the score board. Analysis of the score variation yields a video summary that consists of semantically important and interesting scenes such as reversal or pursuit. Experimental results indicate that the proposed method can summarize basketball video with reasonable accuracy.

Fast affine transform for real-time machine vision applications

In this paper, we have proposed a fast affine transform method for real-time machine vision applications. Inspection of parts by machine vision requires accurate, fast, reliable, and consistent operations, where the transform of visual images plays an important role. Image transform is generally expensive in computation for real-time applications. For example, a transform including rotation and scaling would require four multiplications and four additions per pixel, which is going to be a great burden to process a large image. Our proposed method reduces the complexity substantially by removing four multiplications per pixel, which exploits the relationship between two neighboring pixels. In addition, this paper shows that the affine transform can be performed by fixed point operations with marginal error. Two interpolation methods are also tried on top of the proposed method in order to test the feasibility of fixed point operations. Experimental results indicated that the proposed algorithm was about six times faster than conventional ones without any interpolation and five times faster with bilinear interpolation.

A System Concept for 3D Human Flow Management Based on 3D Spatial Information

abstract A controller who is responsible for safety makes decisions concerning measures for visitor safety using a human-based decision-making process. Many potential accidents that could be caused by human error lurk in the process. These accidents can be reduced by changing from a human-based to technology based decision-making process. A technology-based decision-making process can draw a controller’s attention to trouble spots by data filtering, alarm filtering, and so on. The controller can thus obtain information on the occurrence of possible accidents proactively. The objective of this study is to suggest a system for 3D human flow management using a technology-based decision-making process. The system is divided into two parts: a 3D spatial information system and a human motion detection system. The former is built from 3D CAD and a geographic information system, and the latter is developed from image processing. This study shows the direction of system development and the method of application, and describes the contribution of the system.

Development of a Prototype for Real-Time Indoor Spatial Information Management

The concept and range of spatial information are gradually expanding to various fields through fusion in a point of time when a ubiquitous environment is rapidly developing. Thus, a prototype for construction of a real-time spatial information management system has been developed by integrating a human movement sensing system into a SICS(Spatial Information Control System) engine that can integrally manage inside spatial information extracted from 3D CAD and outside spatial information of GIS. Also, the future research development direction was decided through a pilot test. The Add-on program was developed to extract spatial information necessary for the SICS engine from 3D CAD information, and a human movement sensing system was developed through human movement sensing technology to obtain information of the number of indoor personnel, and thus the interface capable of being utilized was developed by integrating these technologies.