Please or Register to create posts and topics.

Setting Control For Terrestrial Scanning? Looking for advice

#1 · September 26, 2024, 2:17 pm

Quote from bc-surveyor on September 26, 2024, 2:17 pm
Are there any terrestrial LiDAR gurus here? I'm exploring the topic of control density for terrestrial scanning (video will be released at the end of this) and so far I've run two separate tests that have really highlighted how little I know on the subject.

I'll briefly touch on the two test sites (3rd is being processed as I type) and my initial findings & hopefully someone that knows much more than me can chime in and point me in the direction of some literature I can start digging into.

All 3 sites were scanned with an RTC & processed with Leica Register 360

1. 30,000 sq ft. industrial site with tons of geometry that should help with cloud to cloud registration, 150 scans very tightly packed together to capture a very dense scan with few shadows (used to create a LOD400 model). I set 20 B&W targets, 12 were used as control, 8 as independent checks. Average residuals on independent checks were 0.009' in all 3 axis combined. I then created a whole new project with no control, only cloud to cloud registration and compared inversed distances between extracted coordinates of the targets from the point cloud to inversed distances between total station least squares adjusted coordinates. Average residual in all 3 axis of 10 distance checks was 0.008'. What this was telling me is that relative accuracy was not increased by adding control in this data set. But, small site, a ton of scans, only one data set so nothing I can hang a hat on. Off to test 2....

2. 350,000 sq ft, parking garage over 3 floors, 349 scans. Same methodology of not using control points, only cloud to cloud registration and same method of comparing inversed distances. Far from ideal scanning conditions, repetitive surfaces, top floor had little geometry (no roof), a ton of trees around, moving floor from passing by cars, people and cars moving between scans, rushed & spaced out scans (scanned all 350k sq ft in 10 hours solo). The results... an average distance residual of 0.017' horizontally & 0.024' vertically over 300' distances, going through floors as well (1st floor CPs inversed to 3rd floor CPs on the other side of the site). This was very surprising to me. I was expecting much worse results. Typically for a site like this we run anywhere from 20-40 CPs, which from these two tests seems like we are going overboard.

3. Third test will be about 150k sq ft between 3 seperate building with poor scan connections between buildings. I am guessing when I inverse between CPs from one building to another I will see large errors. Time will tell...

Thoughts? Opinions?

*Disclaimer: I am in no way suggesting control & check shots are not needed for scanning, I'm just trying to figure out the optimal density of control for terrestrial scanning jobs. I am currently reprocessing all 3 sites with varying levels of control points added in to see how these numbers change with more control being applied.

Are there any terrestrial LiDAR gurus here? I'm exploring the topic of control density for terrestrial scanning (video will be released at the end of this) and so far I've run two separate tests that have really highlighted how little I know on the subject.

I'll briefly touch on the two test sites (3rd is being processed as I type) and my initial findings & hopefully someone that knows much more than me can chime in and point me in the direction of some literature I can start digging into.

All 3 sites were scanned with an RTC & processed with Leica Register 360

1. 30,000 sq ft. industrial site with tons of geometry that should help with cloud to cloud registration, 150 scans very tightly packed together to capture a very dense scan with few shadows (used to create a LOD400 model). I set 20 B&W targets, 12 were used as control, 8 as independent checks. Average residuals on independent checks were 0.009' in all 3 axis combined. I then created a whole new project with no control, only cloud to cloud registration and compared inversed distances between extracted coordinates of the targets from the point cloud to inversed distances between total station least squares adjusted coordinates. Average residual in all 3 axis of 10 distance checks was 0.008'. What this was telling me is that relative accuracy was not increased by adding control in this data set. But, small site, a ton of scans, only one data set so nothing I can hang a hat on. Off to test 2....

2. 350,000 sq ft, parking garage over 3 floors, 349 scans. Same methodology of not using control points, only cloud to cloud registration and same method of comparing inversed distances. Far from ideal scanning conditions, repetitive surfaces, top floor had little geometry (no roof), a ton of trees around, moving floor from passing by cars, people and cars moving between scans, rushed & spaced out scans (scanned all 350k sq ft in 10 hours solo). The results... an average distance residual of 0.017' horizontally & 0.024' vertically over 300' distances, going through floors as well (1st floor CPs inversed to 3rd floor CPs on the other side of the site). This was very surprising to me. I was expecting much worse results. Typically for a site like this we run anywhere from 20-40 CPs, which from these two tests seems like we are going overboard.

3. Third test will be about 150k sq ft between 3 seperate building with poor scan connections between buildings. I am guessing when I inverse between CPs from one building to another I will see large errors. Time will tell...

Thoughts? Opinions?

*Disclaimer: I am in no way suggesting control & check shots are not needed for scanning, I'm just trying to figure out the optimal density of control for terrestrial scanning jobs. I am currently reprocessing all 3 sites with varying levels of control points added in to see how these numbers change with more control being applied.