|
A study on occlusion removal technology using integral imaging with semantic segmentation and predictive labeling. |
|
Hyun-Woo Kim, Jong-Hoon Huh, Min-Chul Lee(Kyushu Institute of Technology, Japan) |
| In this paper, we propose an improved occluded object visualization method using semantic segmentation and predictive labeling based on integral imaging (InIm) technology. InIm is a passive 3D visualization technique that can generate 3D information through elemental images with different perspectives of information about an object and, it can be used to remove occlusion. However, occlusion pixels in elemental images can degrade the quality of reconstructed images. Our proposed method can remove occlusion and visualize the object of interest. In addition, we propose a predictive labeling method for reducing processing time, which can generate accurate 3D object information without occlusion. |
|
|
Improvement of depth mapping and occlusion removal method using integral imaging |
|
Haruka Sakanoue, Hyun-Woo Kim(Kyushu Institute of Technology, Japan), Myungjin Cho(Hankyong National University, Korea), Min-Chul Lee(Kyushu Institute of Technology, Japan) |
| In this paper, we propose the depth mapping and occlusion removal method using integral imaging (InIm). InIm can remove occlusion using parallax obtained by photographing an object from multiple viewpoints. However, occlusion removal results in an overall darkening of the image. This paper aims to improve occlusion removal using the matrix of the mask image to calculate the number of overlaps. It is difficult to provide depth mapping object due to the influence of the occlusion. To solve this problem, we propose a method of depth mapping and occlusion removal method using InIm. Our method is expected to make great progress in improving safety in the field of autonomous driving. |
|
|
Noise removal in scattering media enviorment using Peplography and DCLGAN |
|
Riki Numata, Hyun-Woo Kim, Seiya Ono(Kyushu Institute of Technology, Japan), Myungjin Cho(Hankyong National University, Korea), Min-Chul Lee(Kyushu Institute of Technology, Japan) |
| Noise removal is important task in reality. The purpose of this research is to remove the noise under the scattering medium environment. As a conventional method, there is a technique called peplography. However, it has a problem that complex noise exists after removing deep noise. We propose a new technology that combines peplography with DCLGAN. DCLGAN is an image transformation algorithm using deep learning which is also used in the field of noise reduction. In this study, we evaluate the conventional method and the proposed method for scattering media images. We believe that the proposed method can be applied in various fields such as object visualization under turbid water. |
|
|
Writing test with image processing technique |
|
Anuntapat Anuntachai(King Mongkut's Institute of Technology Ladkrabang, Thailand), Natte Raksadawan, Kamolchanok Charoy-boon, Nattakan Kruesang(King Mongkut’s Institute of Technology Ladkrabang, Thailand) |
| Nowadays, Dysgraphia frequently discovers as same as other intellectual disabilities. The dysgraphia diagnosis method is to do monitoring by human then this paper represents the applied technology for improving the diagnosis performance. The writing test is to estimate the writing ability and alphabet perception. The image processing is applied to evaluate writing test for students (grade 1-3 or 6-8 years). The student wrote sentences accordingly to an example article in a time frame. The example article composed of 4 lines, 40 words and 161 alphabets of Thai language. |
|
|
A method for estimating Time To Collision from local expansion rate of moving images obtained directly using the spatio-temporal differentiation |
|
Ryotaro Morikawa, Teruo Yamaguchi(Kumamoto University, Japan) |
| This study attempts to estimate the Time To Collision (TTC), the time between camera and obstacle collisions, using only a monocular camera. The TTC is estimated from the expansion rate of the moving image. A spatio-temporal differential method used in this study estimates the expansion rate directly, assuming constant motion parameters within a local model of the image, instead of analysing velocity vectors. The results show that the estimated TTC decreases by about 1.25 s for every 1 s decrease in the actual TTC, although there is an offset between the estimated and actual TTC. |
|
|
Edge Deployment of Vision-based Model for Human Following Robot |
|
Sumaira Manzoor, Eun-jin Kim, Sang-Hyeon Bae, Tae-Yong Kuc(Sungkyunkwan University, Korea) |
| We present edge deploymentment of AI-based vision models: SPHDT & SPBDT, for human-following robot by combining the latest advances of deep learning (DL) and metric learning (MrL). For both models, we leverage DL-based single object detector MobileNetSSD with MrL-based re-identification model, DaSiamRPN and perform qualitative analysis considering six major environmental factors: pose change, illumination variations, partial occlusion, full occlusion, wall corner, & different viewing angles. We select SPBDT model for tracking on the robot and use it to extract target's relative position, location, distance, and angle to control the robot's movement for performing the human-following task. |
|