Improving the Accuracy of Mobile LiDAR for Engineering Surveys

A 3.268Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

During the past two decades, significant progress has been made in laser ranging technology, which is well known as LiDAR named after Light Detection and Ranging. The outcome of various commercial products has marked its turning point from theoretical and laboratory research to application-oriented research and industry development. LiDAR has been introduced to Geomatics Engineering as a novel technique since 2000. Especially in the past few years, a great variety of applications-oriented case studies have progressively been advancing this technology in term of its practicability, efficiency, and performance. Their attractive achievements have created great potential to meet the high demand of LiDAR technology in engineering applications from traditional digital surface/elevation model generation to 3D modeling products for urban planning, transportation, structure inspection, monitoring and engineering design.

In comparison with airborne LiDAR technology, the terrestrial mobile LiDAR system possesses a high potential to meet the high accuracy requirement of Engineering Surveying in terrestrial surveys. It takes full advantage of the most recent technological advances in the Global Position System (GPS), Inertial Navigation, laser scanning, digital imaging, and data processing methods and software. It has been demonstrated that terrestrial mobile LiDAR systems can be superior to the traditional surveying instruments with respect to, for example, reduction of lane closures, risk reduction of casualties, and higher productivity in emergency management practice.

In addition to its use for three-dimensional (3D) modeling in mapping, the terrestrial mobile LiDAR technology has evidently shown great potential to meet the high accuracy requirement in engineering surveying in general. However, to make it qualified as a standard method for engineering surveys in harmony with other existing instruments in a complementary way, more effort is demanded to improve the positioning accuracy of LiDAR points and standardize the procedures for mobile LiDAR from mission planning, data acquisition, and processing to the end products.

For the past several years, we have been carrying out studies to evaluate and improve the performance of the terrestrial mobile LiDAR system, with the support of Ministry of Transportation of Ontario under its Highway Infrastructure Innovation Funding Program, and with Optech Inc. Based on our assessments the point clouds as the end product acquired by Optech’s Lynx system are able to achieve an absolute accuracy of 4.4 cm vertically and 6.0 cm horizontally, and a relative accuracy within 3.65 cm at the 95% confidence level without the aid of any ground control points.

In order to improve the accuracies of mobile LiDAR systems to achieve the high accuracy required by engineering surveying standards, we are conducting a comprehensive analysis on the error budget of a mobile LiDAR system and exploiting the use of linear and planar features to improve positioning accuracies of LiDAR measurements. In this article, we will share some of the preliminary results.

Like with any surveying instrument, LiDAR measurements are subject to errors caused by various factors; but unlike others, the accuracy of a mobile LiDAR system depends not only on the uncertainties associated with the instrument itself, but also on the reliability, stability and accuracy of the vehicle trajectory determined by the on-board GNSS-aided inertial navigation system (INS) functioning as the direct georeferencing system. Additionally, certain systematic errors may exist if the lever arms and boresight angles between the INS and LiDAR instrument can only be calibrated to a limited accuracy level.

Based on error propagation theory, we calculated the errors in the 3-D positions of LiDAR points shown in Figure 1 due to the uncertainties in three groups of factors: INS positions (X, Y, Z coordinates), INS orientations (roll, pitch, and heading); LiDAR measurements (vertical and horizontal angles, ranging distance, and zero error); the lever-arms between the INS sensor and LiDAR instrument (X, Y, Z coordinates), and the boresight angles between them (roll, pitch and heading). Two scenarios were simulated, one as the ideal case having the best calibrations of the last two groups of factors and the other with relative low calibration accuracies. The used uncertainties in the simulations are listed in Table 1. For both cases, the uncertainties associated with the vehicle trajectory (the first group of the factors) were extracted from the Applanix POS SBET solution under a condition with a good GPS coverage. The simulation results are shown in Figures 2 and 3.

As expected, the errors in the LiDAR 3D points caused by the uncertainties in the vehicle trajectory were the largest in the best scenario with accurate calibration parameters; and the errors resulted from the uncertainties in the calibration parameters were increased for the case with less accurate calibrations. It is also noted from Figures 2 and 3 that the errors caused by the uncertainties in the INS orientations, boresight angles, and LiDAR measurements were not uniform across the whole scene with large errors occurring in the areas above the road surface.

To account for the interactions in term of the effects of various error factors on LiDAR positioning accuracies, a global sensitivity study was carried out for the same data set (in Figure 1) and with the same two scenarios (Table 1). The first-order Sobol index for each group of factors was calculated; it indicated the fraction of the variance in a given 3D LiDAR point that could be explained by the variance in each group of factors. That is to say, the contribution of the uncertainties in each group of factors to those in the LiDAR 3-D points can be reflected by the Sobol index. The results are shown in Figures 4 and 5.

The following points can be deducted from Figures 4 and 5 for the simulated scene. (1) The contributions of the uncertainties in the trajectory positions and LiDAR measurements to the errors in the LiDAR 3-D positions were higher than any other factors in the case having the best calibration parameters. (2) The accuracies of the boresight angles and lever-arms between the INS and the LiDAR instrument were dominant factors determining the accuracy of the final solutions for the scenario with less accurate calibration parameters. (3) The contributions of each group of factors were not uniform across the whole scene except for the INS orientations, which made sense for such a short simulated straight distance.

The results from the above-mentioned error analysis indicated that accurate calibration of the LiDAR instrument and the lever-arms and boresight angles is essential for obtaining high accurate LiDAR 3D positions. Errors in the LiDAR solutions may not be uniform for a given scene. Thus, caution should be taken in any further refinement of the LiDAR measurements. The 3-D conformal transformation commonly used to improve the LiDAR positioning accuracy with the aid of ground control points may not be sufficient or valid for a large stretch of data collected with nonlinear vehicles trajectory.

In our studies, we also implemented different strategies to achieve the accuracy improvement of terrestrial mobile LiDAR systems by effectively using ground control points (GCPs). They included: (1) the traditional 3-D conformal transformation with GCPs; (2) 3-D conformal transformation based on GCPs and with linear and planar features as constraints; and (3) 3-D conformal transformation using a limited number of GCPs with linear and planar features as constraints. The developed methods were tested using the data collected by an Optech’s Lynx Mapper (Figure 5). The accuracies of the original LiDAR data were assessed using 22 GCPs on the wall and 6 points on the ground. The Root Mean Square (RMS) errors were 4.25 cm and 4.04 cm horizontally and vertically, respectively. Please note that the calibration parameters were provided by Optech Inc and no further calibration was carried out for this test.

The 3-D conformal transformation based on 18 GCPs was applied to the LiDAR points. The RMS errors calculated based on 10 independent test points were 3.52 cm and 1.65 cm in the horizontal and vertical directions, respectively. The strategy of using 3-D conformal transformation based on the same 18 GCPs and 9 features (6 lines and 3 planes) constructed from these GCPs reduced the horizontal RMS error to 3.32 cm, but the vertical RMS error was increased to 1.90 cm. When the number of GCPs was reduced from 18 to 10, the horizontal and vertical RMS were 3.10 cm and 2.26 cm, respectively. These results showed that incorporating features with GCPs could improve the accuracies of LiDAR measurements. Besides, this strategy could reduce the number of GCPs. The accuracies would be much higher if an accurate calibration were carried out before the 3-D conformal transformation.

In conclusion, promising results were obtained to improve the accuracies of LiDAR data by applying 3-D conformal transformation based on GCPs and features. Tests on large data sets are needed. Piecewise 3-D conformal transformations or non-linear models may be needed based on the non-uniformities of the errors in the LiDAR 3-D points caused by the uncertainties in INS orientations and the boresight angles between the INS and LiDAR instrument.

Drs. Baoxin Hu and Jianguo Wang are professors in the Dept. of Earth and Space Science and Engineering, York University. Michael Leslar and Guannan Liu are graduate students. The research focuses of their group include high precision and high accuracy Engineering Surveying, multi-sensor-aided inertial integrated kinematic positioning and navigation, and 3D characterization of surface objects using optical and LiDAR data.

A 3.268Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE