[image 00799] 屋外で使用可能な近距離用デプスカメラについて / Looking for outdoor depth cameras with short range

[Akihiko Yamaguchi]

Robotics-ML および Image-ML の皆様，
（重複して受け取られた場合はご容赦下さい．）

NAIST/CMU の山口と申します．
私がCMUでお世話になっているChris Atkeson先生からの代理投稿です．

屋外で使用可能な近距離用デプスカメラを探しているのですが，どなたか情報をお持ちの方はご連絡くださいますでしょうか．
詳細は以下の通りです．
Atkeson先生( cga at cmu.edu
)に直接コンタクトを取ってくださっても結構ですし（英語でお願いします），私宛に送ってくださっても結構です（日英どちらでも）．

---------------------------------------------------
Anybody use small depth cameras outdoors at short ranges?

I am looking for a depth camera to be mounted on a robot hand
that works outdoors at ranges up to 50cmm and down to 1cm (ideally).
The goal is to localize objects as the hand approaches them (door
and other handles, railings, and other graspable supports).
Any recommendations?

Here are some things I know about:

Kinect 2 (The new version of the Kinect uses time of flight
technology). $200

Camboard nano http://www.pmdtec.com/products_services/reference_design.php
$700

Camboard pico?
http://www.pmdtec.com/news_media/video/camboard_pico_s_high_framerate.php

Bluetechnix - Argos 3D P100 - ~$1200
http://ww2.bluetechnix.com/en/products/depthsensing/product/argos3d-p100/

Fotonic $4000-$5000
http://www.fotonic.com/assets/documents/products/Fotonic_C-series.pdf
http://www.fotonic.com/assets/documents/products/Fotonic_E-series.pdf

Some comments I have received so far:

- Swiss Ranger did not work well in the DARPA ARM project, even indoors.
Low resolution. It gave weird data due to light reflection.
E.g., a vertical board (e.g., a door) on a table top got distorted, i.e.,
both table and board became "rounded" (like through a fish eye lens). This
created serious localization errors.

- We did some simple experiments with Kinect 2 (which uses time-of-flight)
outside last week. In overcast conditions it seems to do ok, seemingly
similar to indoors operation. In bright sunlight there is notable noise and
the operational range seems to sharply decrease (to roughly 50% of the
original). We just did this very quickly as a proof of concept and were
looking at much longer range operation than what you're targeting ...
- Here is a link discussing kinect 2 in sunlight.  Direct sunlight at a
couple of meters has no depth data.
https://groups.google.com/forum/#!topic/openkinect/GryZfM2vt3M

- We have a camboard nano. Outdoors I don't think you will get 1m. The
Argos 3d might work. It is the same as the camboard nano except with a
stronger light.

- There is also a panasonic 3D depth camera that is supposed to work
outdoors but from what we have seen it is pretty noisy. If budget is
not an issue ASC has some flash lidar and will release an automotive
flash lidar soon (next year sometime).

- One way to reduce the effect of sunlight is to use polarizing filters.
Ideally a system would use multiple polarizing filters or rotate one
and average across multiple measurements with different polarizations, to
reduce the effect of specularities.
http://link.springer.com/chapter/10.1007%2F3-540-61123-1_176
http://www.opticsinfobase.org/ao/abstract.cfm?uri=ao-35-19-3570
http://www.ncbi.nlm.nih.gov/pubmed/21102750
http://filmicgames.com/archives/233
Taking into account polarization can provide additional features for
vision.
CLASSIFICATION OF MATERIAL SURFACES
USING THE POLARIZATION OF SPECULAR
HIGHLIGHTS
academiccommons.columbia.edu/.../CUCS-333-88.pdf
Polarization-Based Material Classification from Specular ...
www.computer.org/csdl/trans/tp/.../i1059.pdf
Separation of Reflection Components Using Color and ...
www1.cs.columbia.edu/CAVE/.../Nayar_IJCV97.pdf
Use heat resistant polarizing filters, you will fry film polarizers.
Edmund Scientific is a source.
Just try them in front of your ToF camera and see what happens.
Experiment to find the best polarization angle. You might want to
continously rotate the polarizer as the ToF camera does its thing.
The best thing to do is build your own ToF camera and take multiple
measurements at different polarizations.

- Structured light approaches that work outdoors:
Observing Optically Challenging Objects with Structured Light
Christoph Mertz
Applied Industrial Optics: Spectroscopy, Imaging and Metrology, July, 2014.

Visual Sensing for Developing Autonomous Behavior in Snake Robots
Hugo Ponte, Max Queenan, Chaohui Gong, Christoph Mertz, Matthew J.
Travers, Florian Enner, Martial Hebert, and Howie Choset
2014 IEEE International Conference on Robotics and Automation (ICRA 2014),
June, 2014.

A low-power structured light sensor for outdoor scene reconstruction and
dominant material identification
Christoph Mertz, Sanjeev J. Koppal, Solomon Sia and Srinivasa G. Narasimhan
Proc. of IEEE Workshop on Projector-Camera Systems (PROCAMS),
June 2012.
http://www.cs.cmu.edu/~ILIM/projects/IL/sunlight/
---------------------------------------------------


どうぞよろしくお願いいたします．
--
/*--------------------------------------------------*-
  山口明彦 (Yamaguchi, Akihiko)
    特任助教,
    奈良先端科学技術大学院大学, 情報科学研究科,
    ロボティクス研究室

    mail : akihiko-y at is.naist.jp
    web  : http://robotics.naist.jp/~akihiko-y/
-*--------------------------------------------------*/