Enhancing 3D Object Detecting for Autonomous Vehicle: Simulation-Based Lidar Data Generatio

acquiring real-world Lidar data for training and testing these models is costly and complex. This project uses simulation environments, specifically the CARLA simulator built on Unreal Engine, to generate synthetic Lidar data and ground truth labels for 3D object detection tasks. The goal is to use simulated data to fine-tune an already trained model for Lidar-based object detection. By simulating various conditions and sensor configurations in CARLA, we aim to capture rich Lidar data along with 3D bounding boxes for cars and pedestrians. This work will evaluate the impact of simulated Lidar data on model performance, comparing results before and after fine-tuning the model with our synthetic dataset. The ultimate aim is to determine whether the integration of simulation-based data can enhance model performance when tested on real-world datasets like KITTI, thereby improving the generalization of AV Lidar-based object detection systems.