context capture: acquisition best practices · data capture requirements and best practices...

1 | WWW.BENTLEY.COM | © 2018 Bentley Systems, Incorporated © 2018 Bentley Systems, Incorporated© 2019 Bentley Systems, Incorporated

Context Capture: Acquisition best practices

Reality Modeling

2 | WWW.BENTLEY.COM | © 2018 Bentley Systems, Incorporated

Content

1. Acquisition best practices

2. Drone mapping acquisition rules

3. Indoor acquisition rules

3 | WWW.BENTLEY.COM | © 2018 Bentley Systems, Incorporated © 2018 Bentley Systems, Incorporated

1 – Acquisition best practices


Acquisition is the key for a successful processing !

Going from failure …. to success !


Best Practices for photo acquisition

Context Capture and photo acquisition:

– Photogrammetric principles and limitations

– Camera model and setup

– Data capture requirements & best practices


Photogrammetric principles

– Based on pixel recognition in multiple photos

– 3 steps process:

• Imagery import

• Aerotriangulation

• Reconstruction


Processing steps

– Aerotriangulation requirements:

• Good Overlap and scene coverage

• Camera calibration: focal length, Principal Point, distortion parameters

• Pose Metadata: Position (X, Y, Z), Orientation (Ω, Φ, Κ)

– Reconstruction requirements:

• Full coverage of the scene

• Appropriate resolution


Photogrammetry limitations

– Limitations:

• Moving objects

• Shiny/transparent/reflective surfaces

• Poor texture / no texture surfaces


Camera model and setup



Hardware:

– Any standard camera (DSLR, Smartphone, bridge, compact, etc.)

– Quality of photo depends of the acquisition device, quality of sensor and optics

Recommendations:

– Keep a constant focal length

– Adjust lighting conditions and camera settings => avoid blurry photos

– Ban geometric retouches

Image format

– High quality JPG, tiff and RAW (RW2, 3FR, DNG, NEF, CRW, CR2)



Shutter speed :

▪ Shutter speed is the most important factor to consider to avoid blurry photos

▪ Easiest parameter to play on



Aperture:

▪ The aperture will affect the amount of light that is captured by the sensor for a given shutterspeed. A small aperture number (f/1.8) means a wide opened diaphragm letting in more light.

▪ The aperture directly affect the depth of field. A bigger aperture number (f/22) means a smallerdiaphragm diameter, letting in less light but with a deeper depth of field, meaning more of thescene in focus.



Camera ISO:

▪ The ISO value represent the sensibility of your sensor. With a high ISO value (high sensibility), yoursensor will need less light for the same pixel brightness. However, a high ISO value will produce morenoise/grain in your photos.


Data capture requirements and best practices

Overlap:

– Smooth angular and distance transition between photos

– At least 60% overlap between consecutive photos

– 15°maximum angle difference between consecutive photos

– Allows the software to find the link between different sides of an object



Coverage:

Homogeneous resolution : same distance to objects within a sequence

To capture more details : get closer + add transition photos

Circle around the scene at different level to reveal any masked areas

Always better to capture too many photos than not enough!



- keep useless information out of the frame: sky, water…

- Necessary to adjust your distance from the subject and/or your optics


Real case examples

Ground survey (Bridge)


Real case examples

Drone 3D mapping (Bentley Systems office)


Real case examples

City mapping (Marseille)


Terrestrial scale project

Stage 1: Mission planning

– Is the site accessible?

– What are its rough dimensions?

– What is my exact area of interest?

– What would be the right camera choice?



Stage 1: Mission planning

– Site will be accessible from north and free of vehicles

– The site is around 1200sqm and the central part is key

– An estimation of 500-800 images will be required to achieve the mission

– Structure seems high but we should not face backspace issues



Stage 2: On-site review

– First global site visit

– Define exact capture path

– Identify potential difficulties

– Determine exact ground control points or scale constraints position

– Camera settings testing



Stage 3: Imagery capture

– Follow main capture path and close the loop

– Check primary capture is robust and with no breaks. Images displayed at speed will appear like a movie.

– Focus on key areas: in this instance, the under-part of the bridge




– Before leaving the site, validate the checklist:

• Main capture path is robust

• Focus areas are good and linked to main capture path

• All geo-registration and measurement information has been collected

• Do not forget any materials on-site

– Basic principle reminder:Going back on-site is expensiveIf in doubt, capture additional pictures

Start & Closing

point

Dense capture

on angles



Stage 4: Processing

• Copy images to your workstation

• Import images and calibration report

• Define optional positioning constraints: GCPs, scale, etc.

• Submit aerotriangulation to ContextCapture desktop or ContextCapture Cloud Processing Service



Stage 4: Processing



Stage 5: Mission closure

– Review time spent on:

• Capture

• Automatic processing

• Manual processing

– The importance of planning and preparation stages will be clear

– Reviewing helps accurately define the time and associated costs for planning further missions

Processing stage Processing time

Capture

1- Planification 30min

2- Global tour 10min

3- Image shooting 60min

Processing

1- Manual processing 5min

2- Automatic processing 240min

Summary Project analysis 20min

TOTAL 6h05min


Processing time

0min

0min

40min

60min

480min

20min

10h00min

Well managed capture Poorly managed capture



Capture

1- Planification 30min

2- Global tour 10min

3- Image shooting 60min

Processing


2- Automatic processing 240min


TOTAL 6h05min


2 – Drone mapping acquisition rules


Drone model

- Fixed wing UAV

- Good autonomy

- Can cover large areas quickly

- Nadir photos only

- Multicopters

- Limited autonomy

- 3D flights allowing for complex flights around assets

- Takes photos from many angles, including upward.


Drone model:

- Drone selection:

- Acquisition quality depends on drone stability and camera quality

- Recommended to use professional survey drones

- Middle range professional drones:

- Integrated camera

- Average sensor size

- Example: DJI Phantom 4

- High end drones:

- High quality DSLR

- High quality DGNSS positioning

- Example: DJI Matrice and Topcon Falcon 8


Drone model: Resolution, Precision and Accuracy

Several factors will affect the survey accuracy and precision:

➢ Quality of the capture device

➢ Quality of the photos

➢ Resolution of the photos

➢ Quality of the geo-referencing

➢ Acquisition pattern

➢ Capability of identifying single features in the scene


Drone model: Resolution, Precision and Accuracy

- Entry price drone and camera

- Georeferencing using GPS tags

- Simple flight pattern (nadir only)

- Quality drone and camera

- Georeferencing using GPS tags

- More complex flight (nadir and obliques)

- Entry price drone and camera

- Georeferencing using ground control points

- Simple flight pattern (nadir only)

- Quality drone and camera

- Georeferencing using ground control points or RTK

- More complex flight (nadir and obliques)

1

2

3

4


Drone acquisition rules

- Nadir Only:

- Difficult camera calibration. Not suitable for buildings

- Obliques grid:

- Grid pattern

- Better for urban environment

- Overlapping circles:

- Overlapping obliques circles pattern

- Better for buildings and industrial plants


Drone capture project

Mission purpose

– Produce a reality mesh of Microsoft Redmond site

– Expected accuracy

• Global site: <5cm

– Merge reality mesh with existing lower resolution mesh (larger extents)

– Drone and GCPs approach will be the best choice



Stage 1: Mission planning from the office

– Define area of take-off and landing and obtain flight clearances

– Identify access to site

– Define drone flight path and GCPs positions for the global area

• Choose the hardware

• Define flight height considering <5cm accuracy expectation



Stage 2: On-site review

– Check planned take-off and landing sites

– Check for potential obstacles that could not be foreseen prior to capture

– Set and survey targets for ground control points




– Execute vertical and oblique predefined flight plan

– Fly at constant height

– Avoid flying over the highway



Stage 4: Processing

• Copy your images to your workstation

• Import images and ground control points

• Register GCPs

• Submit aerotriangulation to ContextCapture or ContextCapture Cloud Processing Service



Stage 5: Mission closure

– Review time spent on:

• Capture

• Automatic processing

• Manual processing

– The importance of planning and preparation stages will be clear

– Reviewing helps accurately define the time and associated costs for planning further missions


Capture

1- Planification 1h

2- On-site checking 10min

3- Image shooting 8h00min

Processing


2- Automatic processing 20days 0h 0min


TOTAL 20d 10h 15min


Processing time

0min

0min

6h 00min

20h 00min

30d 0h 0min

20min

31d 2h 20min

Well managed capture Poorly managed capture



Capture

1- Planification 1h

2- On-site checking 10min

3- Image shooting 8h00min

Processing


2- Automatic processing 20d 0h 0min


TOTAL 20d 10h 15min


Which distance to the objects? Ground Sampling Distance (GSD)

• GSD (ground sampling distance) dependson the camera model, the lens and thedistance from the scene

– R: ground sampling distance

– D : distance to the scene

– f : focal length

– Ls : sensor size

– L : image size in pixel Ls x D / f x L

• R =𝐿𝑠 ∗𝐷

𝑓 ∗𝐿

• Expected precision : P = 2 x R

The calculation of the GSD using this formula is accurate for surfaces that are perpendicular to the line of sight.


Calculation of the overlap

• Drone flight planning

– Projected image size on the ground: 85 X 63 meters

– Longitudinal overlap: 80 %

– Side overlap: 60%

– Distance between consecutive photos: 63 X 20/100 = 12.6 m

– Distance between two flightlines: 85 X 40/100 = 34 m

Surveys drones are generally provided with flight planning software that handles this calculations.


Flight planning software

▪ When capturing data with a drone, it is strongly recommended to use a flightplanning application

▪ Some application allows for flight patterns dedicated to Reality Modeling

Example: Drone Harmony (DJI):


3 – Indoor acquisition rules



Interior survey: Challenging task!

- Short distance and numerous objects / features creating mask

- Light conditions and windows

- Necessary to increase the number of photos and sequences to increase the overlap: 80%

- Short focal or Fisheye lens is recommended


Indoor survey - short focal length lens: fisheye

- Non-linear restitution

- Matrix Calibration

- Potential vignetting with bad lighting conditions and average quality lenses


Indoor survey - short focal length lens: aspheric lenses

- Rectified image with aspheric lens: perspective

- Standard linear calibration

- Potential excessive distortions and low keypoint detection in the corners


Indoor survey: aperture priority!

• Keep a high depth of field, narrow aperture -> « aperture priority »

- High ISO: advantage with a good quality camera

- The camera will increase the exposure: 1/40, 1/50 can be used for nearby objects and short focus distance

- Aperture wider than f/5.6 -> blurry objects

• Camera testing before the acquisition:

- ISO limit of the camera

- Auto-focus capacity, reduced with wide apertures


Indoor survey: distance to objects

• Keep a global context

- The algorithm is using surrounding objects for block consolidation

- Transition sequences between areas of interest

- Large sensor size: Keep a high resolution while acquiring a large context

- Changing light condition and windows: monitor images exposure

• ‘Short’ focus distance

- Reduced depth of field

- Potential blurred background


Indoor survey: acquisition environment

• Changing light

- Low light: avoid using flash, better to use a lighting balloon

- ‘Luminous halo’: avoid reflects, downgraded pixel information around the spots

- White balance: automatic or personalized

• Industrial environment

- Smooth surfaces: use of laser scan, temporary painting

- Add QR codes or April & chilly tags to improve image matching

- Dust: usage of UV filter and regular cleaning


Interest of QR code for indoor survey

• Time-saving and efficient for

– 3D registration (User Tie-Point)

– Automatic Geo-Referencing (GCP)

– Asset tagging

• Automatic detection in images

• No GPS geo-referencing indoor

• Can be used as tie-point on smooth/evensurfaces QR code generated by ContextCapture


Use of QR codes in Context Capture

• Generation using dedicated menu

• 3 types:

– User tie point: Robust AT and asset tagging

– Control Point: georeferencing

– Constraint: Origin, scale, axis, plane

• QR code features:

– ‘Quick Response’

– Up to 4300 alphanumeric characters

– Medium redundancy (15% of code can be restored)


Thank you for you attention

context capture: acquisition best practices · data capture requirements and best practices...

Documents