Drone Project
Drone Project
Drone Project
demo and the source code of the platform are available for public access in
https://github.com/whxru/CoUAV.
Hydra Fusion Tools is able to model so quickly because it exploits the latest
generation of graphics processing units (GPUs). These tiny computers were
originally developed to process graphics for visually complex video games.
Recently, however, their ability to handle graphical information has been put
to use in commercial technologies like augmented reality, and now,
photogrammetric processing.
If your computer is equipped with a powerful enough GPU, in other words, Hydra Fusion
can process the video beamed down from your drone and generate 3D maps on the fly.
Your computer may already have the specs necessary.
As for how fast it goes, John Molberg at Lockheed told New Scientist that the
company can already “fly a drone along at 30 knots” and create a map as it
goes. (That’s about 34 miles per hour for us laymen.)
The model you generate at that speed might not be survey-grade (30
centimeters per pixel) but common sense dictates that if you slow the drone
down, you can get denser data. This means that after one flight to gather
coarse data, you could fly the site again to gather more detail on areas of
interest. With a long enough battery life, someday you might be able to do all
of this before bringing your drone in for a single landing.
Today, by using the latest drone mapping software and a little practice, high-
quality aerial maps can be created by any competent drone pilot. These can be
used by a number of industries, including construction, surveying, mining,
building inspection, forest, and agriculture to create high-resolution mapping
and imagery in real time.
This enables farmers to better plan crop rotation, allows insurance companies
So now you know how useful 3D mapping software is and how various
industries can use it, what are the best 3D mapping solutions available for
drone operators? To help you make an informed decision we’ll take a closer
look at the top 5 drone mapping software solutions on the market.
These are the five biggest players and we’ve assessed each platform for its
image quality, ease of use, range of features and drone compatibility. So
without further ado, let’s get started.
1. PIX4D
Pros:
Powerful feature-rich software with a range of versions available to suit
specific industries. Can easily connect with industry-specific
management platforms to help create yield predictions, profit and loss
accounts, planning documents and detailed site plans. Technical support
is also one of the best in the industry.
Cons:
Expensive for the casual user but the range of features available more
than compensates for this. Due to its complexity, the software can be
difficult to master for beginners.
2. CONTEXTCAPTURE
Pros:
It has truly great 3D modelling capabilities and also allows users to
incorporate data from other survey data collection methods. This is a
highly recommended alternative to Pix4D for enterprise level
organisations.
Cons:
This software is quite expensive and it also requires some level of
training and expertise – not as intuitive as Pix4D.
3. AGISOFT
The platform also has support for NIR, RGB, thermal and multi-spectral
imagery so it has all the tools needed to create highly accurate 3D maps
for structural engineers, architects, and GIS professionals.
Pros:
Cost-effective all-in-one software suite with a full range of image sensors
including NIR, RGB, thermal and multi-spectral. The professional version
creates beautiful clear imagery which is as good as more expensive
software. All things considered, this is a well-rounded platform suitable
for most applications.
Cons:
The software feels clunky to use and the support leaves a little to be
desired compared to Pix4D. The one licence, one computer policy may
also be a problem for some larger organisations.
4. DRONE DEPLOY
There are three paid versions available, Pro, Business, and Premier
which provide the ability to measure volumes or group control points and
export data and images to third-party software including AutoCAD,
Blender, and Rhino.
Pros:
DroneDeploy is a great platform for users who are just getting started
with 3D drone mapping software. While the free version is too limited for
most commercial enterprise uses. The paid versions do provide
increased image quality and have the ability to connect with a wide
range of third-party drone accessories and software.
Cons:
Surface detail for buildings is somewhat disappointing compared to the
more specialised solutions from Agisoft and Pix4D. This makes the
platform more suited to agricultural applications than construction or
engineering.
5. PROPELLER NETWORK
Propeller Network is ideal for heavy civil and resource operations and
helps you and your team to work better together to answer questions
about site productivity and progress.
Propeller also has the incredibly handy AeroPoints. They are the world’s
first smart Ground Control Point system and have been specifically
designed for drone surveying. They are lightweight and durable and are
already used in thousands of surveys in some of the toughest
conditions.
Pros: Propeller is ideal for team data collections, and is a quick and
efficient data collection and management system. It also has data
analytics that is tailored for your industry, as well as their AeroPoints
feature.
As you can see, users are spoilt for choice when it comes to choosing
high-quality capable drone mapping software. All the platforms
mentioned here provide a very high level of quality 3D map reproduction
and are compatible with the most popular drone manufacturers including
DJI and Parrot.
In our opinion, the Pix4D platform offers the best all-around performance across
multiple industries. While it is not the cheapest solution it does offer a complete
range of sensors and map options. It can be seamlessly integrated with back-office
systems using the integrated REST API.
The platform is also very intuitive and easy to use with support from Pix4D rated as
second to none.
So if you’re in the market for a powerful 3D mapping solution for your business, you
won’t go far wrong with the Pix4D platform.
Pix4D is the leading UAV photogrammetry software, specifically designed for users
who want to make 3D maps and point clouds from data captured during a drone
flight.
The COPTRZ 1-day Pix4D training workshop aims to give you an overview of Pix4D
and how to use the software to give you the data output that you require. The
workshop is ideal for those with no previous experience using the software.
Introduction to SLAM technology with Intel
realsense
Robots and Drones need to understand their location and surroundings
more accurately than GPS allows in GPS denied environments and
inside yet un-mapped spaces. RealSense SLAM enables such
applications.
RealSense SLAM uses a fisheye camera, accelerometer, gyroscope,
and depth camera to track a system’s movement in 6DoF. Unlike
commonly used alternative systems, this depth-enhanced visual-inertial
SLAM can track 6DoF motion without any prior knowledge of the
environment. It also allows a location that was mapped previously to be
recognized–this is called relocalization. Together, tracking and
relocalization allow robots and drones to build and share knowledge
about an environment so they can work smarter, or interact
collaboratively.
The depth camera in the RealSense SLAM solution allows a robot to
build a 2D occupancy map. This map shows where obstacles are vs.
space that is free for the robot to operate in. Application logic can tag
locations in this map or otherwise assign meaning to them. This allows
the robot to answer the question not only of ‘where am I now’ but also
‘how do I get to my destination’.
RealSense SLAM and Dense Reconstruction uses little enough CPU
that it leaves ample power for application logic and other sensor
processing. By giving a robot or drone situational awareness, the ZR300
camera with RealSense SLAM helps build better solutions, faster.
The binaries for RealSense SLAM are distributed with sample
applications to help you get up and running quickly. You can review the
code and see how to correctly configure SLAM with your ZR300
camera’s calibration, and how to pass sensor data to SLAM. The header
files and derived API documentation contain definitions of data types,
inputs, and output.
LIST OF TERMS
FEATURE LIST
6DOF TRACKING
Provides a camera pose containing yaw/pitch/roll orientation and x,y,z
position of the camera. The camera pose is provided in real-time with
low latency. The right-handed coordinate system is gravity-aligned and
is defined as: Origin located at the projection center of the fisheye
camera, X axis is horizontal where positive is right, Y axis is vertical
where positive is down, Z is horizontal (optical axis of the camera) where
positive is forward. The world origin is the point where tracking first
began, except when a relocalization map is being used. Updates to the
camera pose are typically provided at 30Hz.
RELOCALIZATION
Relocalization allows an area to be “learned.” The system can then
recognize that it has been in a location before, and report a position
relative to a previously-defined coordinate system. Normally, when
SLAM is started, it has no information about where the device is in an
absolute sense. Tracking is relative to the starting point. However, if the
area has been previously learned, a relocalization map can be loaded
that will allow the system to reorient itself right from the beginning.
Relocalization can also be done on-the-fly without loading a database.
The system learns as it goes, and if it returns to an area it has visited
before, it can recognize it and therefore improve the pose estimation. If
there is a difference between where it thought it was and where it
recognizes it is now, it will correct its estimate of where it thinks it is. This
is useful for correcting drift in the pose estimation, and for recovering
from tracking failures. Relocalization is dependent on successful 6DoF
tracking.
2D OCCUPANCY MAPPING
Occupancy mapping is the process of building a 2D map of an area that
contains information about which areas are occupied by obstacles and
which are open. This can be useful for a robot that needs to know where
it can safely navigate. The occupancy map is a grid where each grid
square has a scalar value indicating the confidence that it is occupied.
The physical size of the grid squares is configurable. The accuracy of
the occupancy map is dependent on the accuracy of the 6DoF tracking.
STARTING SLAM
When SLAM first begins, the camera must be motionless, and there
must be sufficient visual features in view. If the camera is pointed at a
blank wall, ceiling, floor, or objects that are close enough to block the
camera’s view, SLAM may not initialize properly.
CPU UTILIZATION
If the CPU becomes overloaded, sensor samples streaming from the
camera may be dropped in order to keep up. This will have a negative
impact on the accuracy of the SLAM output.
RELOCALIZATION
Relocalization can be done on-the-fly, or with a previously generated
map of landmarks. The system can learn an area, then generate a map
to be used in a future visit. If a relocalization map is available, it should
be loaded before starting SLAM.
SLAM DEVELOPMENT
From a high level, using the RealSense SLAM library is a matter of continuously
feeding camera and IMU data into the module, and continuously receiving the output
of the module as a camera pose and occupancy map. Additionally, the module can
produce files for relocalization and the occupancy map.