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

Area of Science:

• Geomatics Engineering
• Robotics
• Sensor Fusion

Background:

• Mobile mapping systems (MMS) utilizing LiDAR technology are crucial for generating accurate point clouds in surveying.
• Microelectromechanical System (MEMS)-based Inertial Navigation System/Global Navigation Satellite System (INS/GNSS) are key for georeferencing in MMS but suffer from noise and bias instability.
• Existing denoising methods like Long Short-Term Memory (LSTM)-Recurrent Neural Network (RNN) have limitations in addressing MEMS IMU data quality.

Purpose of the Study:

• To develop a low-cost, accurate MEMS-based INS/GNSS georeferencing system for LiDAR mobile mapping.
• To introduce novel denoising and filtering techniques to enhance the accuracy of MEMS IMU data.
• To validate the system's performance against established geospatial data production standards (ASPRS).

Main Methods:

• Development of a wavelet neural network (WNN) for denoising MEMS IMU data.
• Implementation of an optimal maximum likelihood estimator (MLE) for INS/GNSS integration.
• Comparative analysis of the proposed WNN denoising against LSTM-RNN models.
• Accuracy assessment of the integrated system for ground and building mapping scenarios.

Main Results:

• The WNN denoising method improved MEMS-based INS/GNSS integration accuracy by approximately 11%.
• The optimal MLE method achieved ~12% higher accuracy compared to solutions without GNSS.
• The proposed WNN denoising outperformed the LSTM-RNN model.
• Navigation solution accuracy ranged from 1-3 cm for ground mapping and 1-9 cm for building mapping.

Conclusions:

• The developed WNN and optimal MLE methods effectively overcome MEMS IMU noise and bias instability.
• The proposed system achieves high accuracy, meeting ASPRS standards for various mapping applications.
• This research offers a cost-effective and accurate solution for LiDAR mobile mapping georeferencing.