• Journal of Broadcast Engineering
• /
• v.25 no.5
• /
• pp.808-811
• /
• 2020

This paper proposes a method to minimize the error of the Kinect camera sensor by using a random undirected Kalman filter. Kinect cameras, which provide RGB values and depth information, cause nonlinear errors in the sensor, causing problems in various applications such as skeleton detection. Conventional methods have tried to remove errors by using various filtering techniques. However, there is a limit to removing nonlinear noise effectively. Therefore, in this paper, a randomized unscented Kalman filter was applied to predict and update the nonlinear noise characteristics, we next tried to enhance a performance of skeleton detection. The experimental results confirmed that the proposed method is superior to the conventional method in quantitative results and reconstructed images on 3D space.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2015.07a
• /
• pp.518-521
• /
• 2015

컴퓨터 비전이나 감시영상 시스템에서 유동인구 수 측정은 보안 및 경제적 측면에서 중요한 과제이다. 본 논문에서는 RGB센서와 Depth센서를 이용해서 유동 인체를 식별하고 추적하며, 유동인구 수를 측정하는 방법을 제안한다. 제안된 방법은 받아온 영상에서 유동인체를 검출하고, 검출된 인체를 추적한다. 추적된 정보는 이동방향 인식을 하기 위해 사용된다. 중첩에 대한 문제를 해결하기 위해 Merge&Separate 방법과 Kinect Merge&Separate방법을 제안한다. 검출된 유동인체에 대한 정보는 다양한 변수들로 저장된다. 또한 RGB센서와 Depth센서의 장단점을 상호 보완할 수 있는 인체 매칭 시스템을 제안한다. 본 논문의 실험은 자체적으로 제작한 영상을 사용하여 실험을 진행하였다. 실험을 통하여 제안된 유동인구 수 측정 시스템이 유동인구 검출률을 향상시킬 수 있음을 보여준다.

• Journal of Engineering Education Research
• /
• v.19 no.6
• /
• pp.49-56
• /
• 2016

The Human Robot Interface (HRI) based on 3D depth sensor on the docent robot is developed and the localization algorithm based on extended Kalman Filter (EKFLA) are proposed through the capstone design by graduate students in this paper. In addition to this, the performance of the proposed EKFLA is also analyzed. The developed HRI system consists of the route generation and localization algorithm, the user behavior pattern awareness algorithm, the map data generation and building algorithm, the obstacle detection and avoidance algorithm on the robot control modules that control the entire behaviors of the robot. It is confirmed that the improvement ratio of the localization error in EKFLA on the scenarios 1-3 is increased compared with the localization algorithm based on Kalman Filter (KFLA) as 21.96%, 25.81% and 15.03%, respectively.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.5
• /
• pp.637-644
• /
• 2021

In general, in the fields of computer vision, robotics, and augmented reality, the importance of 3D space and 3D object detection and recognition technology has emerged. In particular, since it is possible to acquire RGB images and depth images in real time through an image sensor using Microsoft Kinect method, many changes have been made to object detection, tracking and recognition studies. In this paper, we propose a method to improve the quality of 3D reconstructed images by processing images acquired through a depth-based (RGB-Depth) camera on a multi-view camera system. In this paper, a method of removing noise outside an object by applying a mask acquired from a color image and a method of applying a combined filtering operation to obtain the difference in depth information between pixels inside the object is proposed. Through each experiment result, it was confirmed that the proposed method can effectively remove noise and improve the quality of 3D reconstructed image.

• International Journal of Internet, Broadcasting and Communication
• /
• v.7 no.1
• /
• pp.15-19
• /
• 2015

The depth information of an image is used in a variety of applications including 2D/3D conversion, multi-view extraction, modeling, depth keying, etc. There are various methods to acquire depth information, such as the method to use a stereo camera, the method to use the depth camera of flight time (TOF) method, the method to use 3D modeling software, the method to use 3D scanner and the method to use a structured light just like Microsoft's Kinect. In particular, the depth camera of TOF method measures the distance using infrared light, whereas TOF sensor depends on the sensitivity of optical light of an image sensor (CCD/CMOS). Thus, it is mandatory for the existing image sensors to get an infrared light image by bundling several pixels; these requirements generate a phenomenon to reduce the resolution of an image. This thesis proposed a measure to acquire a high-resolution image through gradual area movement while acquiring a low-resolution image through pixel bundling method. From this measure, one can obtain an effect of acquiring image information in which illumination intensity (lux) and resolution were improved without increasing the performance of an image sensor since the image resolution is not improved as resolving a low-illumination intensity (lux) in accordance with the gradual pixel bundling algorithm.

• The Journal of the Korea Contents Association
• /
• v.16 no.3
• /
• pp.43-51
• /
• 2016

This paper presents the concept of "interactive teleprompter", which provides the user with interaction with oneself or other users for live television broadcasts or smart mirrors. In such interactive applications, eye contacts between the user and the regenerated image or between the user and other persons are important in handling psychological processes or non-verbal communications. Unfortunately, it is not straightforward to address the eye contact issues with conventional combination of normal display and video camera. To address this problem, we propose an "interactive" teleprompter enhanced from conventional teleprompter devices. Our interactive teleprompter can recognize the user's gestures by applying infra-red (IR) depth sensor. This paper also presents test results for a beam splitter which plays a critical role for teleprompter and is designed to handle both visual light for RGB camera and IR for Depth sensor effectively.

• Journal of the Korean Society of Industry Convergence
• /
• v.24 no.6_1
• /
• pp.693-698
• /
• 2021

As technology advances, there is increasing need for research in different fields where this technology is applied. On of the most researched topic in computer vision is Human activity recognition (HAR), which has widely been implemented in various fields which include healthcare, video surveillance and education. We therefore present in this paper a human activity recognition system based on scale and rotation while employing the Kinect depth sensors to obtain the human skeleton joints. In contrast to previous approaches that use joint angles, in this paper we propose that each limb has an angle with the X, Y, Z axes which we employ as feature vectors. The use of the joint angles makes our system scale invariant. We further calculate the body relative direction in the egocentric coordinates in order to provide the rotation invariance. For the system parameters, we employ 8 limbs with their corresponding angles each having the X, Y, Z axes from the coordinate system as feature vectors. The extracted features are finally trained and tested with the Long short term memory (LSTM) Network which gives us an average accuracy of 98.3%.

• Journal of Korea Multimedia Society
• /
• v.15 no.6
• /
• pp.827-836
• /
• 2012

After successful advent of Microsoft's Kinect, many interactive contents that control user's 3D avatar motions in realtime have been created. However, due to the Kinect's intrinsic IR projection problem, users are restricted to face the sensor directly forward and to perform all motions in a standing-still position. These constraints are main reasons that make it almost impossible for the 3D character to navigate the virtual environment, which is one of the most required functionalities in games. This paper proposes a new method that makes 3D character navigate the virtual environment with highly realistic motions. First, in order to find out the user's intention of navigating the virtual environment, the method recognizes walking-in-place motion. Second, the algorithm applies the motion splicing technique which segments the upper and the lower motions of character automatically and then switches the lower motion with pre-processed motion capture data naturally. Since the proposed algorithm can synthesize realistic lower-body walking motion while using motion capture data as well as capturing upper body motion on-line puppetry manner, it allows the 3D character to navigate the virtual environment realistically.

• Journal of Institute of Control, Robotics and Systems
• /
• v.19 no.5
• /
• pp.457-461
• /
• 2013

This paper proposes an image feature-based real-time RGB-D (Red-Green-Blue Depth) 3D SLAM (Simultaneous Localization and Mapping) system. RGB-D data from Kinect style sensors contain a 2D image and per-pixel depth information. 6-DOF (Degree-of-Freedom) visual odometry is obtained through the 3D-RANSAC (RANdom SAmple Consensus) algorithm with 2D image features and depth data. For speed up extraction of features, parallel computation is performed with GPU acceleration. After a feature manager detects a loop closure, a graph-based SLAM algorithm optimizes trajectory of the sensor and builds a 3D point cloud based map.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.2
• /
• pp.763-774
• /
• 2015

Hand posture recognition has been a wide region of applications in Human Computer Interaction and Computer Vision for many years. The problem arises mainly due to the high dexterity of hand and self-occlusions created in the limited view of the camera or illumination variations. To remedy these problems, a hand posture recognition method using 3-D point cloud is proposed to explicitly utilize 3-D information from depth maps in this paper. Firstly, hand region is segmented by a set of depth threshold. Next, hand image normalization will be performed to ensure that the extracted feature descriptors are scale and rotation invariant. By robustly coding and pooling 3-D facets, the proposed descriptor can effectively represent the various hand postures. After that, SVM with Gaussian kernel function is used to address the issue of posture recognition. Experimental results based on posture dataset captured by Kinect sensor (from 1 to 10) demonstrate the effectiveness of the proposed approach and the average recognition rate of our method is over 96%.