Efficient sensor fusion of LiDAR and radar for autonomous vehicles

• Autores: Francisco Javier Méndez Gómez
• Directores de la Tesis: Diego Pedro Morales Santos (dir. tes.) , Manuel Pegalajar Cuéllar (codir. tes.)
• Lectura: En la Universidad de Granada ( España ) en 2022
• Idioma: inglés
• ISBN: 9788411173506
• Tribunal Calificador de la Tesis: Encarnación Castillo Morales (presid.) , Juan Gómez Romero (secret.) , Manuel J. Marín Jiménez (voc.) , Cristina Rubio Escudero (voc.) , Aranzazu Otin Acin (voc.)
• Enlaces
  • Tesis en acceso abierto en: DIGIBUG
• Resumen

  • Autonomous driving is more relevant recently thanks to the advances achieved by companies such as Tesla or Google. This is a result of the technological advances in the field of artificial intelligence at the same time as new automotive sensors are been developed or further optimized, such as LiDAR sensors, which are highly relevant for autonomous driving tasks. However, since it is assumed in the future there will be a large number of autonomous vehicles in the streets, it must be ensured that these vehicles can perform as expected in all normal scenarios present in the real world and not only in laboratory conditions. Therefore, it is required to integrate sensor fusion techniques in the target detection pipeline for autonomous vehicles.

    This doctoral thesis is the result of a research about technologies for computer perception based on sensor fusion techniques as well as deep learning algorithms for target detection for autonomous vehicle applications. These results are presented in the form of a compendium of publications, compiling the scientific articles published during the doctoral period.

    The research has been planned taking into consideration the constraints related to the network edge regarding device memory and latency expected for real time applications. The general objective has been the implementation of the full target detection pipeline in an edge device, including the data preprocessing as well as the evaluation of the data using artificial intelligence techniques. For this purpose, relevant automotive sensors for perception that can be used for autonomous driving have been researched (such as camera, LiDAR and radar sensors) as well as techniques to preprocess the data provided by these sensors in order to maintain relevant features while reducing their memory size and complexity. After this, deep learning algorithms that can be used for target detection following a sensor fusion paradigm have also been researched. Lastly, these algorithms have been optimized to fit in edge devices, such as the Google Coral TPU used in this research.

    The research has been carried out under a doctoral contract at the facilities of Infineon Technologies AG, at its headquarters in Munich, Germany.