Object detection techniques based on deep learning achieve remarkable success. In order to train deep learning-based object detection models, large amounts of labeled data are required. The study was motivated by the fact that the required 3D data can be generated completely in the simulation environment. Within the scope of the study, an algorithm is presented for systematic data collection. The presented algorithm is implemented using ROS and Gazebo with Real Sense D435 camera configurations. Consequently, R GB image, depth map, and point cloud data aligned with the RGB image were captured. In addition, 2D/3D labels are automatically generated for the objects in the recorded data. The obtained data were visualized and thus validated. Sample deep-learning models were trained using the generated data, and the results were presented within the scope of the study. Initial models that are to be trained with actual data can be trained using synthetic data generated based on the obtained results.