abstract: A non-line-of-sight technique is presented to estimate the position and temperature of an object occluded from a camera via reflection on a wall. Because objects emit far-infrared light according to their temperature, their positions and temperatures can be estimated from reflections on a wall. Our key idea is that light paths from a hidden object to the camera depend on the position of the hidden object. The position of the object is recovered from the angular distribution of specular and diffuse reflection components while the temperature of the heat source is recovered from the estimated position and the intensity of reflection. The effectiveness of our method is evaluated in real-world experiments. It is seen that the 2d position of hidden object can be tracked from the distribution of reflection on the wall using a thermal camera.
language of the presentation: Japanese
abstract: Material classification technology is an important element in scene recognition. However, it is difficult to identify the material only with RGB images. In recent years, research on classification using temporal response of light has been conducted, but it is difficult to observe a very fast temporal response of light with a conventional camera. In this study, we realized temporal response observation by high temporal resolution measurement using SPAD sensor, and performed classification using machine learning. The model was evaluated by segmenting multiple materials in the scene.
language of the presentation: Japanese
abstract: It takes a long time to scan a whole scene for capturing of a hyperspectral image with thousands of spectral channels. Any temporal variation in the environmental illumination is therefore a problem, especially when the system captures light passively. We propose a method for compensating temporal inconsistency of environmental illumination on whisk-broom hyperspectral imaging. In addition to the standard 2d scan, we specifically introduce an extra 1d scan orthogonal to the scan line as a reference for the compensation. Our method utilizes a low dimensional structure in the spectral property of illumination to robustly compensate for the illumination transition. We quantitatively evaluate the compensation methods through an experiment under a controllable illumination environment and present an experiment captured stained glass in a cathedral under natural sunlight.
language of the presentation: Japanese
発表題目: 機械走査式ハイパースペクトルイメージングにおける環境光変動補償
発表概要: ハイパースペクトルイメージング技術は農作物、森林、土壌のモニタリングや化学物質のイメージングなどの様々な分野で活用されている。 しかし、詳細な解析のための高精度な撮影には時間がかかり、環境照明の時間的変動が問題となる。そこで本研究では、機械走査式ハイパースペクトルイメージングにおける環境光変動を補償する方法を提案する。 提案手法では、環境光のスペクトル特性における低次元構造を利用することで環境光変動に対するロバストな補償が可能である。 有効性を確認するため、制御可能な照明環境下での実験を通して提案手法を定量的に評価する。また、実環境での応用例として実際の文化遺産のステンドグラスの計測実験において提案手法による補償を行う。
-->
Abstract:
Conventionally, we have been measured the optical response from the scene for understanding the scene.
In recent years, the optical response captured from the scene has been used in the field of computer vision research. For example, see the object out of the line of sight, recognize the material, and reconstruct the object shape using the spectral information.
These studies measure temporal responses in the order of several picoseconds and spectral responses in the order of nanometers.
However, an expensive and special device is required to measure the temporal and spectral response. In general, there is a limit to the improvement in hardware performance, therefore the current measurement method has a limit in improving the resolution.
In this study, we propose a method of measuring optical response with high resolution than conventional methods by encoding and decoding different from the conventional system.
The proposed method achieves high-resolution measurement in two dimensions, temporal and spectral axis that is impossible with a general device devising hardware coding and software decoding.