When DJI first made the M300 and RC Plus compatible I was almost shocked. For the first time we saw a drone upgrade it's Ocusync version with just a firmware update. Not just one generation of Ocusync, it went from V1 to V3 just by firmware! Yesterday's L2 announcement finally showed why this was done. The standard RC that has always come with the M300 doesn't have the CPU power to process the L2 data. So the L2 would be exclusive to the M350 had this not been done. DJI customers are told over and over that RC's and drones that we know could work together simply cannot. We now see that it actually can easily be done if it affects DJI's bottom line. At the end of yesterday's presentation I asked point blank when the M3E would be compatible with the RC Plus. There was a few seconds of silence followed by a quick "there are no current plans". If it could carry an L2 I guarantee it'd already be compatible.

Yup. This is the (DJI) Way.

Nothing surprising from DJI.  We are starting to see promotional videos made by DJI and by the companies that DJI has selected to test the equipment.  It's good, it's more powerful, more precise... They take the time to explain to us that L1 was crap...
So far I don't see any comments from testers who have the honesty to  say that the LiDAR is delivered without post-processing software.  Only DJI Terra which is responsible for generating the cloud point.  Oh yes !  now there is the option to classify ground points and generate DEM.  The classification is fine, although for the price of the license we have the right to expect better.  The DEM... take a sheet of toilet paper, crumple it, it will have the same effect.
From my experience the sensor does 20% of the work, the rest is the software.

Furthermore, to return to the subject, we are told that we must acquire the RC Plus to be able to continue using the M300 in the case of acquiring this LiDAR.  The reason as explained by John is a matter of the power of the RC.  There are some interesting features, yes.  Like the preview of the point clouds present on the SD card.  With L1 you can only have this insight during the flight, during the execution of the mission.  There is also a telemeter/pinpoint function, why not.  Nothing special. These are things that we largely do without when we do mapping.

There are reviews where we try to explain to you that DJI is stronger than Riegl for example.  We'll talk about it again when they provide the software.


I'm waiting to receive a dataset, but I'm not expecting anything, just a denser point cloud.  As expressed above, if there is nothing behind DJI Terra, it is unusable anyway.

I also hope they made an effort to reduce or eliminate positioning errors.  At a minimum, we are warned during the flight that the base station is too far away when using RTK. That would be a good feature.

LV_Forestry Posted at 10-12 10:43
Nothing surprising from DJI.  We are starting to see promotional videos made by DJI and by the companies that DJI has selected to test the equipment.  It's good, it's more powerful, more precise... They take the time to explain to us that L1 was crap...
So far I don't see any comments from testers who have the honesty to  say that the LiDAR is delivered without post-processing software.  Only DJI Terra which is responsible for generating the cloud point.  Oh yes !  now there is the option to classify ground points and generate DEM.  The classification is fine, although for the price of the license we have the right to expect better.  The DEM... take a sheet of toilet paper, crumple it, it will have the same effect.
From my experience the sensor does 20% of the work, the rest is the software.

The L2 is still unavailable even to dealers at the moment. I assume DJI handpicked the few that have their hands on one for obvious reasons. This DJI sensor has a decent price and that's the only advantage it holds to the rest of the industry. Anyone who takes LiDAR seriously isn't going to even give the L2 much thought. I have processed a couple datasets they provided. I'll attach a screenshot from one of them. It is hilarious that this flat, featureless place would be used to showcase a new product like this.

Not to opine on the L2 merits or lack thereof, but a flat, featureless place is ideal for assessing the noise band of a sensor. Just sayin'.

patiam Posted at 10-13 07:55
Not to opine on the L2 merits or lack thereof, but a flat, featureless place is ideal for assessing the noise band of a sensor. Just sayin'.

Yes, but that's the whole problem. We have to go through Terra to obtain the point cloud in LAS format. Terra takes a little liberty with the reality on the ground depending on the settings chosen. Simply click on “smooth point cloud”, which is a paid function, to completely destroy a point cloud:


In red it's the "real" ground, in blue it's the fantasy of Terra when you use the paid version.

What is interesting about LiDAR is to see the fidelity of the repetitions between each pass. I'll give you the example of roofs because it's the most obvious. On trees it's worse but you have to have the eye to detect it. In the example below it is the same dataset, two different passes, on the roof gutter we can clearly see that there is an offset problem, on the ground it is much less obvious. The offset is purely XY, on Z it is very well aligned.



We always come back to the same observation, DJI makes very good hardware, but in terms of software they are not good.

So I come back to this flat surface, the usefulness of such a dataset as a demo is very limited. Especially since no file with GCPs is provided to allow us to make a possible comparison.

Afterwards, we must recognize that DJI seems to be making an effort. In previous versions of Terra it was necessary to have a base station relatively close to the site, otherwise the reconstruction of the dataset was simply refused without any possibility of making the data usable. With version 3.9 I note that this problem is largely reduced. Of course with the distance from the base station the precision of the point cloud is not the best, but at least the dataset is usable.

I was speaking to the utility of multiple passes over a featureless, flat area for use in assessing accuracy & precision. Conceptually, and of course assuming the software used for processing and analysis honors the original data and refrains from "fantasy" smoothing, interpolation, plane-fitting, and the like.

There's a place for those types of data "massaging". and assessing data quality is not it.

Johnnokomis Posted at 2023-10-12 13:39
The L2 is still unavailable even to dealers at the moment. I assume DJI handpicked the few that have their hands on one for obvious reasons. This DJI sensor has a decent price and that's the only advantage it holds to the rest of the industry. Anyone who takes LiDAR seriously isn't going to even give the L2 much thought. I have processed a couple datasets they provided. I'll attach a screenshot from one of them. It is hilarious that this flat, featureless place would be used to showcase a new product like this.
[view_image]

The L1 and L2 are not very good on their own. Combined with good control and post processing the L1 gets considerably better. I only just got the L2 and am very interested in discovering if we can get the precision even greater.

LV_Forestry Posted at 2023-10-17 07:08
Yes, but that's the whole problem. We have to go through Terra to obtain the point cloud in LAS format. Terra takes a little liberty with the reality on the ground depending on the settings chosen. Simply click on “smooth point cloud”, which is a paid function, to completely destroy a point cloud:
[view_image]

I replied earlier without having read the entire thread yet. I personally don't do the processing, I do the data collection and take an interest in the processing done by my supervisor, but it's he that takes our data and increases the precision. Something I hope to learn more about because data is only good in so many applications without the work he does.