The development of autonomous vehicle technology has driven research in the field of localization and mapping, where the use of sensors such as cameras and Light Detection and Ranging (LiDAR) is becoming increasingly common. The combination of data from these two sensors provides richer information for building accurate three-dimensional environment maps. However, the main challenge is to improve localization accuracy, especially under complex and dynamic conditions. Deep learning methods based on Convolutional Neural Networks (CNN) have been widely used to process sensor data, but there is still room for improvement in improving localization accuracy.
This research aims to improve localization accuracy by modifying the CNN architecture used to process camera and LiDAR data. The Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) Odometry dataset is used as the basis to test the performance of the model. In this study, several modifications were made to the original CNN, includi