DJI L1 LiDAR Initial Review and Data comparison

Does the DJI L1 match up to the GVI LiAir V70?
The DJI L1 and V70 occupy a similar end of the LiDAR market, and that’s not where the similarities end. They are both compatible with the DJI M300/skyport and the DJI D-RTK2 Base station, and both can be deployed using DJI’s Mission Planning Software. Not only this, but they also share the same LiDAR sensor, the Livox Avia, which is a low-cost frame sensor capable of three returns, with 70° field of view, and up to 720,000 triple return points per second. 
With so many similarities, are they practically the same product? We put them back-to-back to determine if they perform as their specifications suggest. 

The mission:
We flew both sensors over a 7ha residential/forested area using the same flight plan; 5m/s and ~80m AGL. We set up a DJI D-RTK-2 GNSS over a known survey mark with good sky visibility, and used the data from this base to process both datasets. A GCP survey was undertaken over the mapped area on the same horizontal and vertical coordinate origin as the GNSS base location. We’ve collected lidar datasets from this coordinate origin before, making it a useful baseline for comparison. Note the DJI L1 data was collected using the RTK coordinates derived in the field, while the V70’s gnss coordinates were post processed in GVI’s LiGeoreference software. 

Both data sets were flown with the exact same mission plan from DJI Pilot app and processed using the same Base station data.

Processing the data:
The L1 data was processed in DJI Terra, which has recently been updated to allow for simple LiDAR processing workflows. This workflow essentially consists of selecting the correct file directory for the downloaded L1 data and pressing “go”. The L1 Base station data is automatically imported into the L1 data.

 
The V70 data was processed in LiGeoreference, which involves the steps of: 

1. Establishing the base coordinates by converting the D-RTK2 base station data into a RINEX format.
2. Post-processing the V70’s trajectory
3. Strip Alignment

Normally we would also undertake outlier removal and remove overlap in liGeoreference. Neither of these options are available in DJI terra so we did not use them in this comparison. 

So how does the data compare?
Both datasets were processed with only the first return; DJI is releasing an update to allow 3 returns soon, while the V70 required a firmware update to give meaningful 3rd returns. 

The DJI L1 data visually appears to be significantly noisier than that from the V70. This could be partly due to there being no option in DJI terra to remove outlier points. Both datasets have comparable point densities. 

Typical roofline from our dataset, 0.1m slice. L1 data is green, V70 is red. 

Road cross section - 0.1m slice. RGB is L1 Data, red is V70 data. 

Intensity view: Left - DJI L1, Right, GVI V70. 

Classification for comparison
Both datasets were classified in GVI’s Lidar360 software to derive ground points using the same standard settings. 
 
Accuracy - is one better than the other?
No*. While the point cloud created by the L1 and processed in DJI terra visually appeared significantly noisier, the comparison of classified ground points to our GCP’s told a different story. Compared to 20 points on hard exposed surfaces, the L1 and V70 returned vertical  standard deviations of 5.5cm and 5.4cm from the GCPs respectively; a statistically insignificant difference. As a check, we ran a dataset from our GVI LiAir S220 through the same comparison and returned a standard deviation of 4.4cm. 

*We will be undertaking more rigorous testing on both datasets in due course to compare other metrics.