Need LiDar Drone Feedback

[PalmettoAerial]

Hello -

Drone owner since 2017. I have been a hobby photographer and in real estate development for about a decade. So it’s been great to merge these two passions over the last few years.

I’d like to diversify a bit and do some topo work for clients and competitors on the side. Lidar is a necessity in the DD process and I would like to be able to offer this service.

Looking for any and all feedback on hardware and software. Best practices and any advice.

Thanks!

 

[Felix le Chat]

Hello -

Drone owner since 2017. I have been a hobby photographer and in real estate development for about a decade. So it’s been great to merge these two passions over the last few years.

I’d like to diversify a bit and do some topo work for clients and competitors on the side. Lidar is a necessity in the DD process and I would like to be able to offer this service.

Looking for any and all feedback on hardware and software. Best practices and any advice.

Thanks!

Before you put a large number of very expensive eggs into the LIDAR basket, take a hard look at what can be achieved using standard photography, double-grid sweeps and SFM (Structure From Motion).
Having used both, the detail from last-return LIDAR and low&slow SFM are surprisingly close. Apart from that, with SFM, you also get a photo-detailed 3D environment render if you've got a desktop that can handle the processing workout.

 

[MavicFlyer2]

Before you put a large number of very expensive eggs into the LIDAR basket, take a hard look at what can be achieved using standard photography, double-grid sweeps and SFM (Structure From Motion).

Hello...I've used our Mavic 2 Pro for mapping a little and have done ortho's and such while using Pix4D software to processes the images. It's been years since I've done this, but it is the first time I heard about LIDAR which has always intrigued me. Also I do 360° Panorama photo's stitching them together using Hugin as the software of choice.
This might be off subject but...Doing the 360° has sparked a curiosity of applying it to interior photography of homes. Your comment about SFM I've never heard before. Can you explain this a little for me ?
Thanks in advance !

 

[PalmettoAerial]

Before you put a large number of very expensive eggs into the LIDAR basket, take a hard look at what can be achieved using standard photography, double-grid sweeps and SFM (Structure From Motion).
Having used both, the detail from last-return LIDAR and low&slow SFM are surprisingly close. Apart from that, with SFM, you also get a photo-detailed 3D environment render if you've got a desktop that can handle the processing workout.

Do you think double grid sweeps would be accurate enough for commercial applications?

 

[Felix le Chat]

Hello...I've used our Mavic 2 Pro for mapping a little and have done ortho's and such while using Pix4D software to processes the images. It's been years since I've done this, but it is the first time I heard about LIDAR which has always intrigued me. Also I do 360° Panorama photo's stitching them together using Hugin as the software of choice.
This might be off subject but...Doing the 360° has sparked a curiosity of applying it to interior photography of homes. Your comment about SFM I've never heard before. Can you explain this a little for me ?
Thanks in advance !

As the drone is in motion capturing video or still images, each frame in relation to the one before and the one after (or all those around it if the motion is complex - a low-level double grid sweep, for example) displays something called motion parallax which provides depth information. 3D mapping software such as PIX4D or 3Df Zephyr create highly complex depth related point clouds from the motion parallax present in the series of root images. This is similar to what LIDAR does.

On the other hand, LIDAR sends laser pulses from the puck (camera) and the time it takes for one pulse to return to the detector in the scanner determines the height of the object that bounced the laser back up to the scanner in relation to the other pulse returns.

Keeping it simple: LIDAR might consist of a cycle of 8 'returns' (there are many, many more). First return will determine the highest point of any physical objects directly beneath the puck, in this case: the tree canopy. The second pulse might miss the first return object and get bounced back by objects lower down, the third pulse lower still and so on until you get to the 8th pulse which misses all the obstacles (leaves, branches, grass) and hits the ground.

The 8th pulse would be the Last Return and the topographical data provided by the Last Return sweep is sparse (because the preceding 7 returns have mapped all the objects ABOVE the terrain, creating the DSM or Digital Surface Map). It is the Last Return that creates the DTM (Digital Terrain Map).

Most people are under the impression that LIDAR is some wonder-technology that can penetrate foliage and show things hidden, this is not the case. LIDAR pulses are thrown for a loop by chlorophyll (leaves), water, glass, asphalt and a load of other hard or semi-hard objects and will skip rendering them completely.

SFM will create complex point cloud data for all levels of landscape detail but will do so and provide photo-realistic detail and textures at the same time.

LIDAR will work regardless of ambient light. SFM needs good photographic conditions.



To quote the Wikipedia article related to this:

"...Structure-from-motion photogrammetry with multi-view stereo provides hyperscale landform models using images acquired from a range of digital cameras and optionally a network of ground control points. The technique is not limited in temporal frequency and can provide point cloud data comparable in density and accuracy to those generated by terrestrial and airborne laser scanning (LIDAR) at a fraction of the cost..."

If you're interested in going down this particular rabbit-hole, the Wikipedia article is a very good starting point.

 

[Felix le Chat]

Do you think double grid sweeps would be accurate enough for commercial applications?

My own opinion? Low & slow...yes.

 

[MavicFlyer2]

As the drone is in motion capturing video or still images, each frame in relation to the one before and the one after (or all those around it if the motion is complex - a low-level double grid sweep, for example) displays something called motion parallax which provides depth information. 3D mapping software such as PIX4D or 3Df Zephyr create highly complex depth related point clouds from the motion parallax present in the series of root images. This is similar to what LIDAR does.

On the other hand, LIDAR sends laser pulses from the puck (camera) and the time it takes for one pulse to return to the detector in the scanner determines the height of the object that bounced the laser back up to the scanner in relation to the other pulse returns.

Keeping it simple: LIDAR might consist of a cycle of 8 'returns' (there are many, many more). First return will determine the highest point of any physical objects directly beneath the puck, in this case: the tree canopy. The second pulse might miss the first return object and get bounced back by objects lower down, the third pulse lower still and so on until you get to the 8th pulse which misses all the obstacles (leaves, branches, grass) and hits the ground.

The 8th pulse would be the Last Return and the topographical data provided by the Last Return sweep is sparse (because the preceding 7 returns have mapped all the objects ABOVE the terrain, creating the DSM or Digital Surface Map). It is the Last Return that creates the DTM (Digital Terrain Map).

Most people are under the impression that LIDAR is some wonder-technology that can penetrate foliage and show things hidden, this is not the case. LIDAR pulses are thrown for a loop by chlorophyll (leaves), water, glass, asphalt and a load of other hard or semi-hard objects and will skip rendering them completely.

SFM will create complex point cloud data for all levels of landscape detail but will do so and provide photo-realistic detail and textures at the same time.

LIDAR will work regardless of ambient light. SFM needs good photographic conditions.



To quote the Wikipedia article related to this:

"...Structure-from-motion photogrammetry with multi-view stereo provides hyperscale landform models using images acquired from a range of digital cameras and optionally a network of ground control points. The technique is not limited in temporal frequency and can provide point cloud data comparable in density and accuracy to those generated by terrestrial and airborne laser scanning (LIDAR) at a fraction of the cost..."

If you're interested in going down this particular rabbit-hole, the Wikipedia article is a very good starting point.

Thank you for your response !

 

[PalmettoAerial]

As the drone is in motion capturing video or still images, each frame in relation to the one before and the one after (or all those around it if the motion is complex - a low-level double grid sweep, for example) displays something called motion parallax which provides depth information. 3D mapping software such as PIX4D or 3Df Zephyr create highly complex depth related point clouds from the motion parallax present in the series of root images. This is similar to what LIDAR does.

On the other hand, LIDAR sends laser pulses from the puck (camera) and the time it takes for one pulse to return to the detector in the scanner determines the height of the object that bounced the laser back up to the scanner in relation to the other pulse returns.

Keeping it simple: LIDAR might consist of a cycle of 8 'returns' (there are many, many more). First return will determine the highest point of any physical objects directly beneath the puck, in this case: the tree canopy. The second pulse might miss the first return object and get bounced back by objects lower down, the third pulse lower still and so on until you get to the 8th pulse which misses all the obstacles (leaves, branches, grass) and hits the ground.

The 8th pulse would be the Last Return and the topographical data provided by the Last Return sweep is sparse (because the preceding 7 returns have mapped all the objects ABOVE the terrain, creating the DSM or Digital Surface Map). It is the Last Return that creates the DTM (Digital Terrain Map).

Most people are under the impression that LIDAR is some wonder-technology that can penetrate foliage and show things hidden, this is not the case. LIDAR pulses are thrown for a loop by chlorophyll (leaves), water, glass, asphalt and a load of other hard or semi-hard objects and will skip rendering them completely.

SFM will create complex point cloud data for all levels of landscape detail but will do so and provide photo-realistic detail and textures at the same time.

LIDAR will work regardless of ambient light. SFM needs good photographic conditions.



To quote the Wikipedia article related to this:

"...Structure-from-motion photogrammetry with multi-view stereo provides hyperscale landform models using images acquired from a range of digital cameras and optionally a network of ground control points. The technique is not limited in temporal frequency and can provide point cloud data comparable in density and accuracy to those generated by terrestrial and airborne laser scanning (LIDAR) at a fraction of the cost..."

If you're interested in going down this particular rabbit-hole, the Wikipedia article is a very good starting point.

This is really great info. Thank you for taking the time to share this.

To take this one step further , I would assume there needs to be a post processing stage where all of this data is repackaged to “look like” the terrain so the developer can use / evaluate the site. And if so, does it differ in appearance etc? What softwares do you like for this application?

Also, the photos that are being taken to be used,
I would assume that’s done from the existing camera the drone already has? (ie there is no need for additional expensive equipment?)

 

[Felix le Chat]

This is really great info. Thank you for taking the time to share this.

To take this one step further , I would assume there needs to be a post processing stage where all of this data is repackaged to “look like” the terrain so the developer can use / evaluate the site. And if so, does it differ in appearance etc? What softwares do you like for this application?

Also, the photos that are being taken to be used,
I would assume that’s done from the existing camera the drone already has? (ie there is no need for additional expensive equipment?)

Any drone camera will do the job, but to be honest, the better ones carry the 20mp sensor (P4Pv2/P4P-RTK/Mavic 2 Pro). I think you might need to consider the older drones because they are the ones that are more likely to have SDK support for 3rd party apps - something the latest offerings sadly lack.

As far as software is concerned, there are many packages that will do a very good job, but the best will be the ones you have to pay for. Even 3Df Zephyr, which will generate startlingly good 3D mesh-maps for free (as long as you don't try to process more than 12 photographs) weighs in at about 3 grand per annum for the unlimited pro-package.

DJI TERRA - in conjunction with... you guessed it... DJI drones such as the Phantom P4 RTK used with the D-RTK base station for greater geopositioning accuracy. The P4P-RTK is capable of achieving 1cm+1ppm RTK/HPA (Horizontal Positioning Accuracy), and 1.5cm+1ppm RTK/VPA (Vertical Positioning Accuracy). The 3D mesh-maps have a tendency to be more accurate compared to other softwares. With the bigger DJI birds (M600) you can invest in dedicated camera modules specifically designed for both photogrammetry (P1) and LIDAR (L1)... but this hardware is EXPENSIVE and so are the camera/gimbal modules.

Esri Site Scan for ArcGIS
This provides a complete drone mapping workflow, comprising of flight planning, image processing and data analysis. You can publish drone maps to ArcGIS Online and ArcGIS Enterprise.

Drone Deploy
Not specifically aimed at the professional but still packs a respectable punch with real-time mapping, autonomous flights and 360° virtual walkthrough mode. It is also compatible with a wide range of third-party apps (which can be a really BIG bonus).

Pix4D
Very versatile - wears many hats - Pix4Dmapper (professional drone mapping). Pix4Dsurvey (a hybrid between photogrammetry and CAD). Pix4Dmatic (large-scale photogrammetry). Pix4Dinspect (industry inspection). Pix4Dfields (crop analysis) Pix4Dreact (First Responder). Also very expensive for the non-basic package. There is also a free app-based alternative that supports non-DJI drones such as Parrot and Yuneec but you have to pay for a cloud subscription to create 2D/3D maps and models.

Agisoft
Photogrammetry and 3D spatial mesh-maps in the basic package but the Pro package offers multispectral/thermal processing and georeferenced DSM/DTM generation.

Then you have WebODM (OpenDroneMap) which is open source that has to be installed manually via command line, but there is a surprisingly cheap paid version that has a windows based installer and GUI. This free package can be used to generate mesh-map point clouds, maps, DEM's (Digital Elevation Models) and 3D models from drone images.

If you don't want to use cloud-based processing (something almost all packages push REALLY hard) you need a very high spec graphics/gaming PC with as much RAM as you can cram onto the motherboard, a blisteringly fast multi-core processor and a top notch GPU card otherwise your machine will either crash when it gets to the final render stage, or hang interminably.