Affordable 3D cameras for ROS developers

April 10, 2020

by

Szymon Dzwończyk

Most of the time when robot developers talk about SLAM they address localisation and mapping with use of Lidar and stereo-vision cameras. In this article we’ll focus on the latter.


We didn’t name the article ‘stereo vision cameras for your rovers’ on purpose as there are at least three different approaches to the problem of vision-based depth mapping: stereo vision cameras, time-of-flight sensors and structured light cameras. With those readings onboard your rover can calculate its position in 3-dimensional workspace knowing not only how obstacles look like, but as well how far and how big they are.

Let's just quickly go through differences between the three approaches and then we'll list different of-the-shelf components that are easy to implement to your ROS-based rovers.


Stereo vision

Stereo vision is the most intuitive concept to understand. It mimics the natural vision perception of human and many other species. Stereo vision camera is essentially a camera that consists of two sensors (two lenses) located at a certain distance one from another. It relies on a difference between the left and right camera images which will be slightly shifted depending on the distance to the object it is looking at. Although the measurement relies on a high-demanding calculation of the images differences, the stereo-vision cameras tend to be the least sensitive to the object material, background or different light conditions and are great to be used for outdoor localisation and mapping.


Structured light

As typically the object you’re looking at is unknown, the biggest issue with stereo vision cameras is you need to rely on image processing of the object contours and distinctive points to take into the depth calculation. It takes a lot of computational power - exactly as in nature. On the other hand, the structured light approach relies on projecting a mesh (most commonly infrared light) to the object and looking for a difference between how it should look like and how it looks like in reality. The bigger the mesh cells the closer the object is and as the mesh curves on the object surface we can as well tell what shape it is. Knowing what to look for, we can calculate with less computing power. On the other hand though, as we rely on projected light reflection, we can’t always be sure how the object’s surface and background light affected the reading.


Time-of-flight

These kind of cameras are widely used in photography and phone cameras. Time-of-flight cameras and sensors rely on a time delay between artificial light release and capture of the light being reflected by an object. As light needs a certain time to travel, the time-of-flight system can calculate distance between camera and the object relying on constant light speed and the delay parameter. The procedure is similar to LiDAR readings, but as the hardware and application is closer to cameras than laser sensors, time-of-flight cameras belong to the list here.


The List

The list is based on our market research and hopefully will be enough for you to choose from. We chose the cameras based on availability and price and evaded the ones that are so specialized that cost a little fortune to get. For sure you’ll find more interesting cameras on the market, but here we’ll focus on the ones you can actually get and work with.


Stereo-vision cameras


Zed.jpeg

Stereolabs ZED / ZED 2 / ZED Mini

Price: $349 - $449

Type: Embedded stereo

Depth Range: 0.3 to 25 m

FOV: 96° H, 54° V

Physical Dimensions: 175x30x33 mm

Interface: USB 3.0

Link to ROS Driver


ECON Tara.jpg

e-Con Systems Tara Stereo Camera

Price: $349

Type: Embedded Stereo Camera

FOV: 60° H

Physical Dimensions: 100x30x35 mm

Interface: USB 3.0

Link to ROS Driver

Notes: Inbuilt IMU


T265.png

Intel® RealSense™ Tracking Camera T265

Price: $199

Type: Stereo camera with tracking hardware and software

Cameras: Two Fisheye lenses with combined 163±5° FOV

Physical Dimensions: 108 mm x 24.5 mm x 12.5 mm

Interface: USB 3.1 Gen 1 Micro B (USB2.0 supported)

Link to ROS Driver

Notes: Integrates wheel odometry



duomc-03.png

duo3d DUO

Price: $595-$695


Type: Stereo Camera

Depth Range: 0.23 m to 2.5 m for for M series

FOV: 170 W with 30 mm Baseline

Physical Dimensions: 57 x 30.5 x 14.7mm

Interface: 480 Mbps USB 2.0 Micro-B

Link to API

Link to ROS Driver

Notes: Pixel size 6 x 6 micrometers. Shutter Speed 0.3 microseconds to 1- seconds. Control Functions: Exposure, Shutter, Brightness. Enclosure 6021 Aircraft Grade Aluminium.


5.png

MYNT EYE

Price: $239-$399

Type: Stereo Camera

Depth Range: 0.5 to 18 m

FOV: 146° D, 122° H, 76° V

Physical Dimensions: 141.9 x 61.5 x 68.4 mm

Interface: USB 3.0

Link to SDK

Notes:Six Axis IMU, 100/200/250/333/500 hz Frequency, IMU&Frame Sync: <1ms



Time-of-flight cameras



Kinect 2.jpeg

Microsoft® Kinect™ 2.0

Price: aftermarket - $50-$300

Type: Time of flight

Depth Range: 0.5 to 4.5 m

FOV: 70° H, 60° V

Physical dimensions: ~250x70x45 mm (head)

Interface: USB 3.0

Link to ROS Driver



Structured light cameras


Asus.jpeg

ASUS® XtionPro™ Live

Price: aftermarket

Type: Structured light

Depth Range: 0.8 to 3.5 m

FOV: 58° H, 45° V

Physical dimensions: ~180x40x25 mm (head)

Interface: USB 2.0

Link to ROS Driver

Notes: Similar internals to the Xbox Kinect 1.0. Discontinued.


D415.jpg

Intel® RealSense™ Cameras

Price: $79 - $199

Type: Active IR Stereo

Depth Range: 0.105 to 10 m /0.3 to 10 m

Depth FOV: 63.4° x 40.4° (+/-3°)

RGB FOB: 69.4° x 42.5° x 77° (+/- 3°)

Physical Dimensions: 99 mm x 20 mm x 23 mm

Interface: USB-C 3.1 Gen 1

Link to ROS Driver


Orbbec® Astra Series

Price: $149.99

Type: Structured Light

Depth Range: 0.6 m to 5.0 m

FOV: 73 D x 60 H x 49.5 V

Physical Dimensions: 80 x 20 x 20 mm

Interface: <2.4w data-preserve-html-node="true" USB

Link to ROS Driver


Want to see more posts like that?

Subscribe to stay informed.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

See more blog posts: