top of page

Laser Scanner Comparison: NavVis Leica RTC360

NavVis – Leica RTC360 system comparison

Area: Seven Dials, London

Date of survey: March 2022 (RTC) - 21. March 2023 (NavVis)

Software used for comparison: CloudCompare V2.13.

Survey control: PGMs and Retro Targets, Traversed and Least Square adjusted using MicroSurvey StarNet

Introduction: Seven Dials is situated in London Borough of Camden. The survey area consists of 3 roads which form a triangle shape: Monmouth Street – Short’s Gardens – Neal Street.


It was surveyed by TopoCrew in March 2022 in order to provide topographical survey data for road design. We opted to use our Leica RTC360 scanner to capture the site.

Having a static scanner comes with its limitations and we have decided to look for a mobile solution as an addition to our ever expending fleet of equipment and one that wouldn’t involve using the Register360 Register software.

The NavVis system seemed to be a logical choice having a body mounted SLAM hardware and a cloud based processing solution called IVION. The available settings and data manipulation seems to be quite limited, but it doesn’t mean that it’s not adequate. It could actually be a very welcomed feature indeed as long as it works as expected. The aim of this study however is to compare the geometry of the point cloud data captured by the two scanners and not the supporting software packages. This being said, since the raw survey data was processed and exported using the appropriate software, the results of this comparison is inevitably affected by the processing software as well.

The two point clouds were reduced close to ground level. They were compared at their geo-referenced position as they were registered by their processing software.

The purpose of the comparison, is to show the relative differences in the topology of the two point clouds and not the absolute deviation between them. Therefor to better understand the true difference between the two data sets, minimising the errors resulting from the geo-referencing / data export, we also transformed the NavVis data onto the RTC360 using a best fit method in CloudCompare allowing only vertical shift of the cloud.

Analysis:

First, the geometry of the two individually geo-referenced point clouds were compared using a cloud-to-cloud distance calculation in CloudCompare with 2D1/2 Triangulation.

The mean error (Fig. 1) was 21mm with the maximum error ranging to about 50mm. These values however shouldn’t be taken as absolute discrepancies because the two solutions fit the control slightly differently and the scene has changed between the two surveys as people and vehicles changed location.

Figure 1 Mean distance errors

Looking at the absolute distances, the fitting error to the control points is apparent as the cloud has a generally green colour which in the applied colour scheme is from approx. 15mm to 35mm.

The red coloured areas are those of interest. There are 3 red areas which seem to be significant enough to be looked at.

The one on Monmouth Street on the West side of the area is by far the largest where we have found significant difference between the RTC360 and the NavVis data.

Figure 2 Error distribution with enhanced colourization


Figure 3 Monmouth Street deviations in the channel aera

Taking a closer look would already suggest what would have caused the discrepancy. The footpath and the central area of the road shows a significantly better fit than the road side (channel) up to the curb line. This wouldn’t be the case if there was a registration error in eighter of the two scan data. Switching the point cloud colour scheme from intensity to RGB clearly shows what had happened. Fig. 4 below shows the street in its recent state during the NavVis data capture, where Fig. 5 shows the same area at time of the first survey, carried out with the RTC360 about a year earlier.

Figure 4 NavVis RGB data – recent survey post road construction


Figure 5 RTC360 RGB data – prior to road construction

The street has since been rebuilt. The difference between the two data was therefor caused by a recent construction activity instead of registration error.

Moving a bit further North on the data, the red area extends out on to the footway. This time, checking the RGB colours would not show any apparent difference between the two data sets and we can confidently assume, that there is an error between the two point clouds.

Figure 6 Area of ongoing road construction and point cloud misalignment


Figure 7 The applied colour scheme shows that the difference is over 30mm

Luckily, there was a control station in the area in question which is considered errorless in this case as it is orders of magnitude more accurate than the point cloud data. After measuring the distance from the control point to the closest point in the point cloud in both data sets, it reveals a shift in both clouds but in the opposite direction. In this unfortunate coincidence, although the measured deviation from the control station is under 20mm in both cases, the resulting difference between the two point clouds is 34mm.

Figure 8 Vertical misalignment of the NavVis data: +18mm


Figure 9 Vertical misalignment of the RTC360 data: -16mm

Although the point cloud is slightly misaligned, the circumstances also need to be taken into consideration. As mentioned earlier there was an ongoing road construction in progress in this area. As a result of this, the possible path around the fenced site was limited to the walkway with many people walking about and criss-crossing the road in order to intersect the trajectory path was not possible.

Figure 10 Fenced area making data capture difficult

The next, less severe area with detected error is on Short’s Gardens. During data capture, a number of lorries were passing through the narrow street, blocking the entire opposite side of the road. This resulted in an uncertainty in the trajectory. It was decided to end the current job and start a new one capturing the data overlapping the previous scene. It is very likely that the weak geometry caused by the traffic and the loose end of the trajectory could have caused this minor error.

Figure 11: Slight misalignment at the join of trajectories, where heavy traffic was disrupting the data capture

The third and last area to mention is not the fault of any of the scans, in fact it clearly shows where works have been taken place between the time the two scans were carried out.

Figure 12 Previous road works detected


Figure 13 Signs of recent roadworks are clearly visible.

For the second attempt, we best fitted the NavVis data onto the RTC360 to be able to see the relative discrepancies between the two clouds minimising the effect of the registration errors.

Figure 14 Calculated distances between the two point clouds best fitted to each other

The distance calculation between the two point clouds shows that 85% of the points are within 20mm. These results should be read and understood with the consideration of the circumstances. There was a full year gap between the two surveys. People and vehicles had a different distribution across the site and even the built environment has changed slightly because of the road works being carried out in the one year time frame.

As a conclusion it is fair to say that to do a direct comparison between the two point clouds, the test area should have been a controlled environment with no changes in the scene, and considering this fact the results show a very strong correlation between the two data sets. Taking into consideration that the NavVis data was collected in about 5th of the time that it took with the RTC360, it is quite remarkable. However, it should also be noted that the NavVis system needs a best practice to be strictly followed to result in good quality data. This was apparent where the construction site didn’t make it possible to intersect the trajectory path in loops.

Finally, there are many more aspects to look at for a full comparison. These could be data density, data spread and noise, or to test the systems in different environment etc. This study however was only limited to understand how the geometry of the two data sets compare to each other.


bottom of page