Optimal Low-Cost MEMS INS/GNSS Integrated Georeferencing Solution for LiDAR Mobile Mapping Applications.

Mobile mapping systems using LiDAR technology are becoming a reliable surveying technique to generate accurate point clouds. Mobile mapping systems integrate several advanced surveying technologies. This research investigated the development of a low-cost, accurate Microelectromechanical System (MEMS)-based INS/GNSS georeferencing system for LiDAR mobile mapping applications, enabling the generation of accurate point clouds. The challenge of using the MEMS IMU is that it is contaminated by high levels of noise and bias instability. To overcome this issue, new denoising and filtering methods were developed using a wavelet neural network (WNN) and an optimal maximum likelihood estimator (MLE) method to achieve an accurate MEMS-based INS/GNSS integration navigation solution for LiDAR mobile mapping applications. Moreover, the final accuracy of the MEMS-based INS/GNSS navigation solution was compared with the ASPRS standards for geospatial data production. It was found that the proposed WNN denoising method improved the MEMS-based INS/GNSS integration accuracy by approximately 11%, and that the optimal MLE method achieved approximately 12% higher accuracy than the forward-only navigation solution without GNSS outages. The proposed WNN denoising outperforms the current state-of-the-art Long Short-Term Memory (LSTM)-Recurrent Neural Network (RNN), or LSTM-RNN, denoising model. Additionally, it was found that, depending on the sensor-object distance, the accuracy of the optimal MLE-based MEMS INS/GNSS navigation solution with WNN denoising ranged from 1 to 3 cm for ground mapping and from 1 to 9 cm for building mapping, which can fulfill the ASPRS standards of classes 1 to 3 and classes 1 to 9 for ground and building mapping cases, respectively.