Is GCP for traffic accident reconstruction necessary


#1

Hi, my unit is in charge of conducting traffic accident reconstruction and we are looking at using drones and photogrammetry such as your solution, Pix4D & Agisoft. I would like to know if GCP is necessary in our field of work, especially when a typical accident scene is < 50 - 70m long. To the forensic world, accuracy of scanning and measurement (point cloud) is important; but in this case of referencing to the map and its GPS coordinates, would it be necessary? I’d like to know how many of the traffic police units or forensic units, when using drones use GCP when scanning and processing photogrammetry data. I’m all ears to the advice from DroneDeploy and the practising community. Thanks in advance.

Andy


#2

Hi @AndyCLk,

Thanks for swinging by the DroneDeploy Forum. I suggest taking a look at our blog post When to Use Ground Control Points as a place to start.

I’m not too familiar with the field of forensics but I hope another member of this community can chime in and help give you a clearer answer.

Cheers,
Christina


#3

Hi Andy,

For your use you would place ground targets, at least 3, spread out with the largest angle between them as practical and measure the relative distances between them. Then when you process your project you can identify your targets and using scale bars in Pix4d you will end up with an image that you can precisely measure horizontal distances.


#4

Hi Christina

Noted with thanks.


#5

Hi Dave, thanks so much for the tips. Would you be able to share other than DronDeploy, what other photogrammetry programs that you had used so far? And your reviews about them too?


#6

You could do what Dave says also in Drone Deploy. Horizontal would likely not be a problem. Relative to within the map, there should not be much of an issue. But sometimes areas tend to bow a bit like a bowl when processed. Usually some sequential overlapping oblique orbital type images added to the upload with the overhead grid can help prevent this bowing. As a record of the site, it is a great asset and performs a very complimentary function. It is in the early stages as far as this application is concerned. I have actually used the 3D Model from Drone Deploy as the basis for a 3D reconstruction animation. I did not use GCP’s and it was very accurate when compared to the survey that had been completed of the site. It depends on the issues of the site and if they are properly accounted for. Experience helps.


#7

Hi Gary,
Thanks for sharing your experience and views. When you mentioned (without GCP) about it being very accurate when compared to survey done, would please help me understand with an elaboration? ie. Was it the point to point (distance or length, etc.) measurement of an object OR some GPS coordinate that you have, as compared to the survey?


#8

Gary,

Are you saying that there is a way to specify relative distance values to targets in Drone Deploy in order to adjust the orthoimage in the x-y axis? Or are you saying that generally, Drone Deploy does a good job with relative accuracy on it’s own?

If it is the latter, then without any user input, measuring horizontal distance in the orthoimage between ground targets of known distance will be accurate to less than 1 inch to prove horizontal accuracy? I have never seen this claim before.


#9

Hi @Dave,

The accuracy within the map is generally accurate to about 1-3% of the resolution of your orhtomasic, this is also called your Ground Sampling Distance (GSD). On a global level, the map is going to be oriented on the basemap with the same accuracy of the GPS on the drone. Depending on a lot of variables, this can be much less accurate. To orient the map correctly on a global level, you would need to use GCPs. Here is a great document on the accuracy of your map.

Cheers!
Zach


#10

Hi @zach1 That’s good to know. Can you provide a case study documenting that high of relative accuracy is typical?


#11

Hi @Dave,

The closest thing we have to a case study about GCPs and accuracy is our Drone Accuracy & Cloud-Based Photogrammetry whitepaper. Hope this is useful.

Cheers,
Christina


#12

Hi Christina,
So far, i could see that DroneDeploy has only cloud-based processing (pls correct me if I’m wrong) for the churning out of 3D images amd possibly point cloud. As we work on cases that are forensically sensitive, we would not want to be put in a position of placing the evidence into a “black box” (with all due respect to tour company’s modus operande) for the necessary processing. Would tou be able to share if we are able to process the photogrammetry images and data in our own standalone processor/laptop/desktop?


#13

@Christina Thank you for the study.

@zach1 According to the study, horizontal accuracy was increased both globally and relatively by a factor of 10 when traditional GCPs were used.

If the operator placed ground targets in the subject area and measured the distances precisely between them, and then DD allowed the user to identify the targets in the images that contain them and input the known distances, I’m guessing the the results would be that relative accuracy would increase by a factor of 10 but would have no effect on global accuracy. That would be a good solution when relative accuracy is all that is necessary on a project without the time or expense of traditional GCPs.

Wouldn’t you agree?


#14

Not sure whether this helps or not but using images obtained using Drone Deploy, they can then be processed in Photogrammetry software such as Agisoft Photoscan. Models can then be georeferenced in QGIS or similar with a level of accuracy that is suitable for archaeological use, and measurements derived from this. What I am less certain of is whether this workflow would be accurate enough for forensic use.


#15

Hi @AvochAndy, thanks for the sharing. Your pointers made me realise that there could be other methods other than GCP (which is resource/expense intensive and time intensive with regard to the necessary strategic GCP markings), in this case, the mentioned QGIS. This made me look into the accuracy of QGIS vs the traditional GPS and how accurate it could be…which then brought me to another topic of figuring out the accuracy of RTK/PPK drones vs the added input of using GCP (see link of Pix4D: https://pix4d.com/rtk-ppk-drones-gcp-comparison/).

In agreement with @Dave… At first thought, local accuracy should be good enough (X and Y coord should be good and relative accurate too) but I understand that the Z-coordinate could be hugely affected without a proper and accurate GPS marking or GCP. Well… depth of vehicle damage and indentations and relative height and position of the damage areas on the body of the vehicles are important. The depth of damage, scratches and gouge marks on the roads would be important too (Z axis again)…hence my reservations now about doing away with GCP. I’ll still try to find out more if such measurements could be accurately taken without GCP…but I’m all ears to further advices.

Well, like I said…being a noob, it leaves me to researching for more and I do hope that I’m on the right track to getting an indication of what drone, positioning accuracy hardware/software and photogrammetry programs to acquire.

Once again, thanks so much for all your help. I’ll be listening in for more :slight_smile:. Have agreat weekend. Cheers!


#16

Hi @AndyCLk,

My colleague Parker Deuel will be reaching out to you via the email address you used to sign up for our forum. He’s interested in hearing about what you’re trying to accomplish and how we can help. :slight_smile:

Cheers,
Christina


#17

Noted Christina. Thanks for the info. I’ll get back to him soon.


#18

Hi. I would be interested in this as well as I have approached local police and have used it for rural Fire investigation.


#19

Hi @Grizz, do you mean that you are working on providing 3d photogrammetry solutions to the fire service and also the local police? Please share (if you could) your experience and how accurate the products of the drone scanning had been. Thank you.


#20

GCP’s aren’t necessary unless you are worried about real world elevations and horizontal tolerances better than 1ft over 1,000ft. I am guessing that the majority of your scenes are pretty small in comparison to what many of us map so no worry about the bowling effect. I would fly in 3D mode though.