Högskolan i Skövde

The rapid advancements in autonomous driving (AD) technologies have significantly emphasized the development of accurate and reliable perception systems, particularly for 3D object detection. This thesis focuses on enhancing 3D object detection in autonomous driving by leveraging LiDAR-Camera fusion. The primary aim is to develop a robust system that integrates the precise distance measurement capabilities of LiDAR with the rich contextual information provided by camera images, thereby improving the accuracy and reliability of object detection in diverse and dynamic driving environments. The objectives of this research include developing a system for sensor fusion, implementing deep learning models to process fused data, and validating the proposed approach through experiments. A pre-trained YOLOv5 model is employed to detect objects in 2D images captured by cameras. The detected objects are then projected into the 3D space using LiDAR data, which is synchronized and calibrated with the camera data. The fusion process involves transforming the LiDAR point clouds into the 2D image plane to associate the depth information with the detected objects, thereby facilitating accurate 3D object positioning. 

The results demonstrate that integrating LiDAR and Camera data improves the accuracy of 3D object detection. The evaluation process, which includes comparing the estimated depths with actual measurements, shows minimal discrepancies, confirming the system’s high accuracy and reliability. This thesis contributes to the field of autonomous driving by providing a validated imelimented system for LiDAR-Camera fusion in 3D object detection. The findings underscore the importance of sensor fusion in enhancing the robustness and accuracy of perception systems in autonomous vehicles. Future work could focus on improving the system’s performance in adverse weather conditions, integrating additional sensors like RADAR, and exploring more advanced deep learning models to further advance the capabilities of autonomous driving technologies.