How High Accuracy Asset Collection Drives Mobile LiDAR System and Extraction Software Requirements

A 4.067Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

As 2014 winds down, we have been taking a closer look at TopoDOT usage statistics for the year. One very interesting development has been the emergence of some real power-users with applications trending toward asset collection in support of Geographical Information Systems (GIS). As TopoDOT was not specifically developed as a GIS application solution, its successful use in this area is intriguing.

As an example of the GIS requirements we focused specifically on a project providing services to the "Highways Agency" in the United Kingdom. The requirements document for this project was "Asset Data Collection Manual Asset Capture Rules" , Edition 1-April 2013 Version: 1.0.

In this article we’ll take a look at some of the requirements of the Highways Agency. Specifically, we will focus on how the ADCM accuracy requirements drove mobile LiDAR system performance and processes designed to extract features with a high level of precision. We’ll discuss how point clouds and images acquired with mobile LiDAR systems must be tightly calibrated to meet these collection requirements. Lastly, a discussion of how operations were successfully organized to accommodate a change in workflow demanded by these relatively higher levels of extraction precision.

ADCM Requirements Driving LiDAR System Performance
LiDAR Data Accuracy

The "Asset Data Collection Manual Asset Capture Rules" (ADCM) document is 165 pages long describing the collection requirements for hundreds of assets. However the key ADCM requirement motivating the need for very tightly calibrated LiDAR data and high precision extraction processes is found on page 7, under "Spatial Referencing". This requirement stated that asset North/East position will be recorded to within the following absolute tolerances:
<300mm for on carriageway features, i.e. road markings, studs and curbs;
<500mm for off carriageway features within 5m of the road edge, i.e. signs, lighting points and crossovers;
<1500mm for off carriageway features beyond 5m of the road edge, i.e. hedges, fences and trees.

The driver on LiDAR system performance is of course the first 300mm (about 1 foot) absolute accuracy requirement. At first glance, this requirement might be well within the performance of modern mobile LiDAR systems. However, it’s safe to say that such projects are acquired without control reference coordinates and most likely use virtual networks for GPS error correction to the trajectory model. Without going into great technical detail, unfavorable GPS and/or reference station baselines could yield significant trajectory drift. Thus one might expect trajectory drift to comprise the majority of the allowable 300mm error budget leaving little extra allowable error in sensor precision, calibration or extraction.

LiDAR Data Relative Accuracy
While the ADCM did not call for any requirements on relative accuracy between objects in close proximity, it is safe to assume that the Highways Agency expects much higher requirements than their standards for absolute accuracy. Simply spoken while they might tolerate up to 300mm error in the overall roadway position, they would expect much higher accuracy for relative distances between assets within the same area. So the entire scene can drift up to 300mm, yet there is a general expectation of parallel paint lines, parallel curbs, as well as a maintaining the relative orientation of assets within the same proximity. Thus one might place a requirement on relative LiDAR data accuracy of one tenth the absolute or about 30mm.

Asset Size
The list of assets described in the ADCM is rather extensive with relatively small objects among those to be collected. One could not expect to recognize these assets within a point cloud alone. Thus a requirement for LiDAR system data containing both a high density accurate point cloud and tightly calibrated high resolution digital images for feature recognition is inferred.

Asset Identification and Extraction
An understanding of ADCM requirements makes clear the need for a "higher-end" mobile LiDAR system featuring high accuracy navigation performance. Clearly the requirement for extracting just under100 different type assets of varying complexity necessitated both a very accurate point cloud for spatial location and tightly calibrated images to assist in recognition.

As shown in the following TopoDOT screen shot image, the importance of calibrated images become clear. This image is indicative of the potential complexity of this data. In this scene several assets are located in close proximity. The previous discussion regarding relative accuracy becomes clear as some assets are within 300m (1 foot) distance from each other.

It should also be clear that high levels of automation in the extraction of these assets would be unrealistic. One could not envision a software algorithm intelligent enough to automatically identify, locate and place the correct cell at the center of each asset. While several software products claim to at least find vertical features automatically, they advertise performance levels of about 70% recognition. Taking them at their word, that still leaves 30% of the assets missed. Sifting through these large data sets in search of these missing or mistaken assets can require more time than a technician identifying them correctly in a semi-automated approach the first time.

TopoDOT’s Asset Identification tool extracts the asset with a semi-automated approach by exploiting the synergy between the point cloud and calibrated image information. This information is provided quickly to the technician with a single mouse click on the asset in either the image or point cloud. As Certainty 3D has nothing against reliable automation, certain asset information is extracted immediately. The diameter and center of a vertical asset is calculated and the selected cell marker is placed on the ground level at that location. Vertical height is also calculated automatically. Orthogonal directions to the selected point cloud are shown so that any tilt can be easily assessed and calculated. Should a mistake be made, the tool allows for easy interactive correction in location and height.

TopoDOT Asset Identification tool also allows the user to define custom lists of associated metadata for each asset. Thus a technician can quickly select relevant metadata from the list. This information is automatically recorded in one of several standard GIS data format files.

The requirements provided in the ADCM document clearly place inherent requirements on the mobile LiDAR system as it pertains to the type of data acquired and both absolute and relative accuracy specifications. The requirement to recognize many different features of varying size along with paint lines, curbs and general roadway topography precluded any the use of any fully automated extraction tools. The semi-automatic approach offered by TopoDOT’s Asset Identification tool proved the most efficient means of extracting the features identified in the ADCM document.

Having gained an understanding of why TopoDOT proved the most productive software solution for such applications, we began to consider the workflow necessary to support these operations. Specifically we looked at the necessity of a rather drastic reorganization of the operational model as compared to operations supporting traditional field intensive technologies.

Reorganization of the Operational Model
Clearly the application of mobile LiDAR technology to GIS applications requires a reorganization of methods to optimize the new operational model. This reorganization can be relatively significant for companies organized primarily around older field technologies. To fully understand the requirement of this reorganization, it is instructive to step back to assess the traditional field intensive process first and then consider how mobile LiDAR technology has altered that process. Then the requirements for reorganization become clear.

Extracting features from data is fundamentally an act of cognitive recognition or more simply stated the "act of recognizing and knowing". Prior to the advent of mobile platform technology, features were simply recognized in the field. A technician would identify an object and determine its location in some way. He would then enter relevant metadata into a database while at the object location. This is typically a very time intensive process with large collections requiring months to years. Yet the data is returned with features already identified and processed. Thus in this operational model downstream processing is then largely focused on integrating this data within a larger GIS application platform.

Mobile LiDAR platform collection technology is extremely fast; so fast that field time is now a negligible part of the overall process. However the key to understanding the implication of these systems with respect to the operational model is that they yield information, i.e. point clouds, calibrated image and related data to which no cognitive recognition is applied. Cognitive recognition is then performed on the data with programs such as TopoDOT. So one must interpret the new process flow as "shifting" the process of cognitive recognition from the field to the office. The overall process is still much faster given the extraordinary increase in field productivity. However office processes must necessarily increase in time.

Attempts to automate these recognition processes are on-going. Interestingly expectations of the level of achievable automation are often driven by past experience of opening up files obtained manually. Despite the now negligible field time, often users still expect downstream feature extraction processes to be at least comparable in time to previous field intensive operations wherein the recognition work was done on-site. That being said, once the new workflow is understood and processing manpower shifted to the office, mobile LiDAR technology applied to GIS applications yields productivity and product quality far exceeding traditional field technology.

TopoDOT is certainly not the primary software application used to process mobile system technology for GIS applications. However it has interestingly proven itself to be the optimal solution for asset collection requirements similar to those of the ADCM document described above. While no formal survey of GIS project requirements was undertaken, we can only surmise that the stringent requirements of the ADCM are still a minority. However, should more GIS project requirements follow such standards, there will be a consequent upgrade in the performance requirements of mobile LiDAR systems acquiring the data and processing procedures that are semi-automatic, efficient and exploit the inherent synergy between point clouds and tightly calibrated images.

Ted Knaak has more than 20 years experience in the LiDAR industry. He founded Certainty 3D LLC, Orlando FL in 2011, and prior to C3D, founded Riegl USA in 1993. He holds MSEE and MSME degrees.

A 4.067Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE