Combustion instability monitoring

My colleagues and I have developed methods for monitoring the combustion instability of a gas turbine used in power plants. I developed a deep-learning-based method taking a sequence of high-speed flame images as input. Our newly proposed early fusion layer is capable of learning to represent the per-pixel power spectral density of the images, which is time invariant under certain conditions. The newly proposed late fusion layer is capable of combining different outputs at different time steps so that the current output will be more accurate.

Related publications

Ouk Choi, Jongwun Choi, Namkeun Kim, and Min Chul Lee, "Combustion instability monitoring through deep-learning-based classification of sequential high-speed flame images," Electronics, vol. 9, no. 5, p. 848, 2020.

Jongwun Choi, Ouk Choi, Min Chul Lee, and Namkeun Kim, "On the observation of the transient behavior of gas turbine combustion instability using the entropy analysis of dynamic pressure," Experimental Thermal and Fluid Science, vol. 115, p. 110099, 2020.

Ouk Choi and Min Chul Lee, "Investigation into the combustion instability of synthetic natural gases using high speed flame images and their proper orthogonal decomposition," International Journal of Hydrogen Energy, vol. 41, no. 45, pp. 20731-20743, 2016.

3D Human Modeling

We have developed a simple yet accurate method for obtaining 3D geometric and photometric models of a human from a single RGB-D image. To this end, we proposed a learning-based method for reconstructing a whole-body point cloud from a single front-view human-depth image. The proposed method is robust to noise and missing data which actual depth images typically suffer from.

Related publications

Jae Won Jang, Young Chan Kwon, Hwasup Lim, and Ouk Choi*, "CNN-based denoising, completion, and prediction of whole-body human-depth images," IEEE Access, vol. 7, no. 1, pp. 175842-175856, 2019.

Young Chan Kwon, Jae Won Jang, Hwasup Lim, and Ouk Choi*, "Feasibility analysis of deep learning-based reality assessment of human back-view images," Electronics, vol. 9, no. 4, p. 656, 2020.

Extrinsic Calibration of Multiple RGB-D Cameras

We are developing a robust, efficient and easy-to-use method for extrinsic calibration of multiple RGB-D cameras. The extrinsic calibration method relies on a monochromatic spherical target, whose projected shape is well-approximated by a circle. We have developed a method for detecting monochromatic circular regions in an RGB image. The method is robust to changes in the illumination or background, while the method is flexible enough to use a basketball as the spherical target.

Related publications

Jae Won Jang, Young Chan Kwon, Wonjun Hwang, and Ouk Choi*, "Robust alternating optimisation for extrinsic calibration of RGB-D cameras," Electronics Letters, vol. 55, no. 18, pp. 992-994, 2019 (interviewed article).

Young Chan Kwon, Jae Won Jang, Youngbae Hwang, and Ouk Choi*, "Multi-cue-based circle detection and its application to robust extrinsic calibration of RGB-D cameras," Sensors, vol. 19, no. 7, p. 1539, 2019.

Young Chan Kwon, Jae Won Jang, and Ouk Choi, "Automatic sphere detection for extrinsic calibration of multiple RGBD cameras," in International Conference on Control, Automation and Systems, pp. 1451-1454, 2018, oral presentation.

Human-projector Interaction

Recent projectors are compact, power-efficient, but bright. Those projectors are getting established in mobile robots and cars. We are developing fingertip-based human-projector interaction methods, which recognize and track fingertips of left or right hands. The following videos show applications of the proposed methods.

Related publications

Ouk Choi*, Young-Jun Son, Hwasup Lim, and Sang Chul Ahn, "Co-recognition of multiple fingertips for tabletop human-projector interaction," IEEE Transactions on Multimedia, vol. 21, no. 6, pp. 1487-1498, 2019.

Ouk Choi, Young-Jun Son, Hwasup Lim, and Sang Chul Ahn, "Cascaded fingertip detection and classification for human-projector interaction on tabletop surfaces," in International Conference on Electronics, Information, and Communication, pp. 1275-1278, 2018.

Young-Jun Son, Ouk Choi, "Image-based hand pose classification using Faster R-CNN," in International Conference on Control, Automation and Systems, pp. 1569-1573, 2017, oral presentation.

Young-Jun Son, Ouk Choi, Hwasup Lim, and Sang Chul Ahn, "Depth-based fingertip detection for human-projector interaction on tabletop surfaces," in IEEE/IEIE International Conference on Consumer Electronics Asia, pp. 189-192, 2016, oral presentation.

Stereo Matching

We developed a stereo matching algorithm that is efficient and robust in the presence of large weakly textured objects in the scene. We hope that the developed algorithm will be applied in real-time systems such as autonomous cars and robots.

Related publications

Ouk Choi and Hyun Sung Chang, "Yet another cost aggregation over models," IEEE Transactions on Image Processing, vol. 25, no. 11, pp. 5397-5410, 2016.

Jaehyuk Choi and Ouk Choi*, "Integrated visual sensor with 2D/3D imaging and in situ proximity sensing for mobile devices," Electronics Letters, vol. 52, no. 16, pp. 1377-1379, 2016.