Creation of a Synthetic Dataset in Blender for CNN-Based Parking Deck Training
Master’s Thesis / Bachelor’s Thesis / Research Internship
Description
The use of synthetic datasets in computer vision is becoming increasingly important, especially for training neural networks. This project aims to create a digital 3D model of a parking deck in Blender to realistically simulate various weather conditions and environmental factors. The generated dataset will serve as the basis for training a Convolutional Neural Network (CNN), which will subsequently be tested and evaluated on real images of the parking deck.
Research Questions
- How realistically can synthetic datasets be generated using Blender?
- To what extent can neural networks trained on synthetic data be transferred to real-world scenarios?
- Which environmental factors have the greatest impact on object recognition in parking decks?
Objectives of the Work
The objective of this project is to create a synthetic dataset that represents various environmental conditions (e.g., rain, fog, shadows) and improves the generalization capability of a CNN. Furthermore, the study will analyze how well the trained network can be applied to real-world images.
Work Packages
- Modeling a 3D parking deck in Blender with realistic structures
- Simulating different environmental conditions (lighting variations, weather effects)
- Rendering images from multiple perspectives to generate the dataset
- Training a CNN using the synthetic dataset
- Testing and evaluating the network on real parking deck images
Your Skills
- Experience in 3D modeling (e.g., Blender)
- Basic knowledge of Deep Learning and Computer Vision
- Programming skills in Python (e.g., TensorFlow, PyTorch)
- Interest in synthetic datasets and image processing
