SensorFusion

SensorFusion using MATLAB

Target

Vehicle attitude, trajectory and lane map reconstruction for sparse feature, GNSS degraded, high speed drive environment

Files

Starting with raw data

• To start with raw data, run 'main.py' file first with appropriate file path (raw data files are not included in this repo)

• 'dataloader.py' file reads '.pkl' format files and extracts data into dictionary variable, and finally converts to '.mat' file

• Modify 'dataloader.py' to extract different raw data from other '.pkl' files

Starting with extracted data

'dataprocessor.m'

• After running 'dataloader.py', there will be various '.mat' format data files

• Running one section in 'main.m' will read the data and process through them for creating dataset of directly usable format

• Timestamps are interpolated, GNSS measurements with poor accuracy are filtered for optimization stability

• Finally, IMU readings are clustered for easier preintegration

'optimizer.m'

• Running the remaining sections of 'main.m' will use 'optimizer.m', which is the main part of this research

• Using Sparse Non-Linear Least Squares optimization algorithms, 'optimizer.m' solves Sensor Fusion problem

• There are currently 3 modes possible, 'basic', 'partial', '2-phase'. These modes are classified based on the sensor data used for optimization

• Read 'optimizer.m' for more detailed explanation