BioMedical Systems Laboratory

Recently, there have been significant advances in self-driving cars, which will play key roles in future intelligent transportation systems. In order for these cars to be successfully deployed on real roads, they must be able to autonomously drive along collision-free paths while obeying traffic laws. In contrast to many existing approaches that use prebuilt maps of roads and traffic signals, we propose algorithms and systems using Unified Map built with various onboard sensors to detect obstacles, other cars, traffic signs, and pedestrians. The proposed map contains not only the information on real obstacles nearby but also traffic signs and pedestrians as virtual obstacles. Using this map, the path planner can efficiently find paths free from collisions while obeying traffic laws. The proposed algorithms were implemented on a commercial vehicle and successfully validated in various environments, including the 2012 Hyundai Autonomous Ground Vehicle Competition.

This paper describes a new methodology for estimating the relative pose between two 2-D lidars. Scanned points of 2-D lidars do not have enough feature information for correspondence matching. For this reason, additional image sensors or artificial landmarks at known locations have been used to find the relative pose. We propose a novel method of estimating the relative pose between 2-D lidars without any additional sensors or artificial landmarks. By scanning two orthogonal planes, we utilize the coplanarity of the scan points on each plane and the orthogonality of the plane normals. Even if we capture planes which are not exactly orthogonal, the method provides good results using nonlinear optimization. Experiments with both synthetic and real data show the validity of the proposed method. We also derive two degenerate cases: one related to plane poses, and the other caused by the relative pose. To the best of our knowledge, this study provides the first solution for the problem.

This paper presents a practical means of extrinsic calibration between a camera and a 2D laser sensor, without overlap. In previous calibration methods, the sensors must be able to see a common geometric structure such as a plane or a line. In order to calibrate a non-overlapping camera laser system, it is necessary to attach an extra sensor, such as a camera or a 3D laser sensor, whose relative poses from both the camera and the 2D laser sensor can be calculated. In this paper, we propose two means of calibrating a non-overlapping camera laser system directly without an extra sensor. For each method, the initial solution of the relative pose between the camera and the 2D laser sensor is computed by adopting a reasonable assumption about geometric structures. This is then refined via non-linear optimization, even if the assumption is not met perfectly. Both simulation results and experiments using actual data show that the proposed methods provide reliable results compared to the ground truth, as well as similar or better results than those provided by conventional methods. (C) 2015 Elsevier B.V. All rights reserved.

The system for recognizing text in natural scenes has been applied in various industries. However, due to the change in brightness that occurs in nature such as light reflection and shadow, the text recognition performance significantly decreases. To solve this problem, we propose an adversarial shadow generation and training algorithm that is robust to shadow changes. The adversarial shadow generation and training algorithm divides the entire image into a total of 9 grids, and adjusts the brightness with 4 trainable parameters for each grid. Finally, training is conducted in a adversarial relationship between the text recognition model and the shaded image generator. As the training progresses, more and more difficult shaded grid combinations occur. When training with this curriculumlearning attitude, we not only showed a performance improvement of more than 3% in the ICDAR2015 public benchmark dataset, but also confirmed that the performance improved when applied to our’s android application text recognition dataset.

오늘날 정확한 노면 인식 시스템 구축을 위해서는 딥러닝 기반의 의미론적 객체분할 기술의 적용이 필수적이다. 하지만 실세계에서 충분히 존재할 수 있는 야지 장면에서 노면인식을 위한 의미론적 객체분할 데이터 셋은 아직까지도 개발되지 않았고, 그렇기 때문에 의미론적 객체분할 기술이 야지 노면 인식 시스템에서 적용된 연구 사례도 매우 적다. 우리는 이러한 문제를 해결하기 위하여 야지 환경에서의 노면 인식 데이터 셋을 구축하고, 지금까지 적용되지 못했던 의미론적 객체분할 기술들을 적용 및 최적하고 분석한다. 실험 결과 우리는 우리의 야지 노면 인식 데이터 셋에서 980FPS의 연산속도로 89.34 mIoU를 달성하였다.

의미론적 영상 분할을 위한 컨볼루션 신경망 기반 접근 방식은 픽셀 단위 레이블을 통한 지도 학습에 의존한다. 하지만 접근 불가능한 도메인으로 일반화되지 않을 수 있는 문제점이 존재한다. 의미론적 영상 분할에서 모든 데이터에 사람이 직접 레이블을 지정하는 작업은 노동 집약적이기 때문에 소량의 라벨링 데이터를 이용해 네트워크의 일반화 성능을 향상시키는 것은 매우 중요하다. 본 논문에서는 가상 도메인과 현실 도메인 사이의 교차 도메인 혼합 샘플링을 활용한 도메인 적응 학습 방법을 제안한다. 이를 위하여 혼합 샘플링 방법의 대표적인 기법을 분석하고 활용하여 도메인 적응 문제에 적용하고 일반화 성능을 비교한다. 제안하는 학습 방법을 통해 학습된 네트워크는 도메인 적응 문제에서 정량적 수치와 정성적 결과 모두 기준 모델을 능가하는 성능을 보인다.