|
A research on scattering removal technology using compact GPU machine for real-time visualization |
|
Seiya Ono, 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 a system that enables visualization under the situation of scattering media by Peplography which is scattering media removal system, on a small GPU machine. The conventional peplography process is performed on a desktop PC, and no system has yet been developed to perform peplography in an outside environment. Thus, we have developed a peplography system using the Jetson AGX Xavier, which is superior to conventional methods in terms of portability, execution speed, and power consumption. We also demonstrate the performance of this system and evaluating the obtained results in terms of FPS and the image evaluation metric. |
|
|
Photon-counting integral imaging using Bayesian estimation with depth camera information |
|
Hideaki Uchino, 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 introduce a method for object visualizing under low light conditions using photon-counting integral imaging and depth prior information. Conventional photon-counting technology statistically estimates photons, enabling 3D visualization under low light via maximum likelihood estimation (MLE). However, MLE ignores prior information, leading to inaccurate results. Also, calculation errors make it difficult to obtain the accurate depth of objects at longer distances. To solve this problem, we propose Bayesian methods like maximum posteriori, using prior data from a depth camera that can acquire information in low light, enhance 3D quality. |
|
|
Enhancement three-dimensional localization of ArUco marker under photon-starved conditions |
|
Jin-Ung Ha, 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 a method to improve the recognition rate of ArUco markers under photon-starved conditions by preprocessing such as photon counting algorithm and histogram equalization to estimate the exact 3D location of the enhanced object. To improve the recognition rate of ArUco markers, photon counting algorithm was used to estimate the photons emitted by ArUco markers, and histogram equalization was used to equalize the photons to remove the random noise generated by photon counting. To verify the feasibility of the proposed method, we implement simulations using the Open Manipulator-X platform and show performance metrics such as PSNR. |
|
|
Utilizing Machine Learning for Predicting the Trajectory of Moving Objects and Implementing a Robot-based Catching Algorithm |
|
Namyeong Lee, Jun Moon(Hanyang university, Korea) |
| Our research focuses on examining the challenge of catching a rolling ball, which is comparatively simpler compared to catching a flying object. However, we present a novel framework that addresses the limitations of traditional methods by solely relying on machine learning techniques. Our approach employs a CNN-based network, enabling the detection of diverse objects. Additionally, we employ a transformer to forecast the trajectory of the object. To ensure efficient training and accurate evaluation in real-world scenarios, we leverage a simulator. Consequently, our framework offers enhanced practicality. |
|
|
Building reliable datasets for autonomous vehicles: Data Factory for the industry and intelligent trains. |
|
Claire NICODEME(SNCF, France) |
| Autonomous vehicles have long been a worldwide research subject. to observe and react in its environment, the vehicle must embed many sensors. Each of them produces huge amount of data. To make sure algorithms and AI-based solutions are properly designed, training/validation/test data must answer some prerequisites. In particular, the industry has more constrains and requirements than academia. This paper proposes a concept of data factory for railway and industrial purposes. First of all, the conception of acquisition plans and data sanity checks are presented. Then storage and necessary volumes are discussed, followed by annotation and access. |
|
|
Method of tracking moving objects combining Kalman filter and K-means |
|
Hitoki Uchida, Teruo Yamaguchi(Kumamoto University, Japan) |
| According to the Ministry of Agriculture, Forestry and Fisheries website, the amount of damage to crops caused by wild vermin was approximately fifteen billion yen in 2021. Of this amount, Eighty percent was caused by animals and Twenty percent by birds. Most of the beasts are wild boars, deer, and monkeys. Therefore, I wondered if it would be possible to take low-cost countermeasures against these three pests, which cause a lot of damage in Japan. Of course, the best countermeasure would be to put up fences or walls around the entire property, but these require a lot of money and a lot of labor when putting them up, which I thought would be difficult for individuals tending to their own far |
|